[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8,"_9":10},"actionData","errors","root",{"_1128":1129,"_1130":1131},"routes/_app.x.$handle.$name",{"_106":105,"_32":51,"_114":96,"_27":115,"_13":14,"_116":-5,"_11":117,"_118":119,"_120":121,"_122":123,"_124":125,"_21":-5,"_126":-5,"_127":-5,"_128":25,"_129":130,"_131":65,"_132":-5,"_133":-5,"_134":135},"routes/_app.x.$handle.$name.branch.$branch.version.$version.diff",{"_11":12,"_13":14,"_15":16,"_17":-5,"_18":-5},"project",{"_49":50,"_32":51,"_38":52,"_53":-5,"_54":25,"_55":56,"_57":56,"_58":59,"_60":-5,"_18":61,"_62":63,"_64":65,"_66":-5,"_67":-5,"_68":69,"_70":71,"_72":73,"_74":75,"_76":77,"_78":-5,"_79":80,"_81":80,"_82":83,"_84":83,"_85":83,"_86":80,"_87":80,"_88":89,"_90":91,"_92":93,"_94":25},"version",653,"changes",[19,20],"aiChatInfo","userId",{"_21":43,"_23":-7,"_24":25,"_26":25,"_27":44,"_29":44},{"_21":22,"_23":-7,"_24":25,"_26":25,"_27":28,"_29":28},"file",{"_30":31,"_32":33,"_34":35,"_36":37,"_38":39,"_40":14,"_41":42,"_24":25},"previous","deleted",false,"deletedHere","path","skills/refs/ref_1.ts","latestPath","fileId","f3604bf2-5d6d-11f1-961a-1607ee4eb77e","name","ref_1.ts","fileType","script","content","// GENERATED by deploy/gen_skills_catalog.mjs — do not edit by hand.\nimport type { SkillRef } from \"../index.ts\";\nexport const REFS_PART_1: SkillRef[] = [\n {\"skill\":\"ab-testing\",\"ref\":\"sample-size-guide\",\"body\":\"# Sample Size Guide\\n\\nReference for calculating sample sizes and test duration.\\n\\n## Contents\\n- Sample Size Fundamentals (required inputs, what these mean)\\n- Sample Size Quick Reference Tables\\n- Duration Calculator (formula, examples, minimum duration rules, maximum duration guidelines)\\n- Online Calculators\\n- Adjusting for Multiple Variants\\n- Common Sample Size Mistakes\\n- When Sample Size Requirements Are Too High\\n- Sequential Testing\\n- Quick Decision Framework\\n\\n## Sample Size Fundamentals\\n\\n### Required Inputs\\n\\n1. **Baseline conversion rate**: Your current rate\\n2. **Minimum detectable effect (MDE)**: Smallest change worth detecting\\n3. **Statistical significance level**: Usually 95% (α = 0.05)\\n4. **Statistical power**: Usually 80% (β = 0.20)\\n\\n### What These Mean\\n\\n**Baseline conversion rate**: If your page converts at 5%, that's your baseline.\\n\\n**MDE (Minimum Detectable Effect)**: The smallest improvement you care about detecting. Set this based on:\\n- Business impact (is a 5% lift meaningful?)\\n- Implementation cost (worth the effort?)\\n- Realistic expectations (what have past tests shown?)\\n\\n**Statistical significance (95%)**: Means there's less than 5% chance the observed difference is due to random chance.\\n\\n**Statistical power (80%)**: Means if there's a real effect of size MDE, you have 80% chance of detecting it.\\n\\n---\\n\\n## Sample Size Quick Reference Tables\\n\\n### Conversion Rate: 1%\\n\\n| Lift to Detect | Sample per Variant | Total Sample |\\n|----------------|-------------------|--------------|\\n| 5% (1% → 1.05%) | 1,500,000 | 3,000,000 |\\n| 10% (1% → 1.1%) | 380,000 | 760,000 |\\n| 20% (1% → 1.2%) | 97,000 | 194,000 |\\n| 50% (1% → 1.5%) | 16,000 | 32,000 |\\n| 100% (1% → 2%) | 4,200 | 8,400 |\\n\\n### Conversion Rate: 3%\\n\\n| Lift to Detect | Sample per Variant | Total Sample |\\n|----------------|-------------------|--------------|\\n| 5% (3% → 3.15%) | 480,000 | 960,000 |\\n| 10% (3% → 3.3%) | 120,000 | 240,000 |\\n| 20% (3% → 3.6%) | 31,000 | 62,000 |\\n| 50% (3% → 4.5%) | 5,200 | 10,400 |\\n| 100% (3% → 6%) | 1,400 | 2,800 |\\n\\n### Conversion Rate: 5%\\n\\n| Lift to Detect | Sample per Variant | Total Sample |\\n|----------------|-------------------|--------------|\\n| 5% (5% → 5.25%) | 280,000 | 560,000 |\\n| 10% (5% → 5.5%) | 72,000 | 144,000 |\\n| 20% (5% → 6%) | 18,000 | 36,000 |\\n| 50% (5% → 7.5%) | 3,100 | 6,200 |\\n| 100% (5% → 10%) | 810 | 1,620 |\\n\\n### Conversion Rate: 10%\\n\\n| Lift to Detect | Sample per Variant | Total Sample |\\n|----------------|-------------------|--------------|\\n| 5% (10% → 10.5%) | 130,000 | 260,000 |\\n| 10% (10% → 11%) | 34,000 | 68,000 |\\n| 20% (10% → 12%) | 8,700 | 17,400 |\\n| 50% (10% → 15%) | 1,500 | 3,000 |\\n| 100% (10% → 20%) | 400 | 800 |\\n\\n### Conversion Rate: 20%\\n\\n| Lift to Detect | Sample per Variant | Total Sample |\\n|----------------|-------------------|--------------|\\n| 5% (20% → 21%) | 60,000 | 120,000 |\\n| 10% (20% → 22%) | 16,000 | 32,000 |\\n| 20% (20% → 24%) | 4,000 | 8,000 |\\n| 50% (20% → 30%) | 700 | 1,400 |\\n| 100% (20% → 40%) | 200 | 400 |\\n\\n---\\n\\n## Duration Calculator\\n\\n### Formula\\n\\n```\\nDuration (days) = (Sample per variant × Number of variants) / (Daily traffic × % exposed)\\n```\\n\\n### Examples\\n\\n**Scenario 1: High-traffic page**\\n- Need: 10,000 per variant (2 variants = 20,000 total)\\n- Daily traffic: 5,000 visitors\\n- 100% exposed to test\\n- Duration: 20,000 / 5,000 = **4 days**\\n\\n**Scenario 2: Medium-traffic page**\\n- Need: 30,000 per variant (60,000 total)\\n- Daily traffic: 2,000 visitors\\n- 100% exposed\\n- Duration: 60,000 / 2,000 = **30 days**\\n\\n**Scenario 3: Low-traffic with partial exposure**\\n- Need: 15,000 per variant (30,000 total)\\n- Daily traffic: 500 visitors\\n- 50% exposed to test\\n- Effective daily: 250\\n- Duration: 30,000 / 250 = **120 days** (too long!)\\n\\n### Minimum Duration Rules\\n\\nEven with sufficient sample size, run tests for at least:\\n- **1 full week**: To capture day-of-week variation\\n- **2 business cycles**: If B2B (weekday vs. weekend patterns)\\n- **Through paydays**: If e-commerce (beginning/end of month)\\n\\n### Maximum Duration Guidelines\\n\\nAvoid running tests longer than 4-8 weeks:\\n- Novelty effects wear off\\n- External factors intervene\\n- Opportunity cost of other tests\\n\\n---\\n\\n## Online Calculators\\n\\n### Recommended Tools\\n\\n**Evan Miller's Calculator**\\nhttps://www.evanmiller.org/ab-testing/sample-size.html\\n- Simple interface\\n- Bookmark-worthy\\n\\n**Optimizely's Calculator**\\nhttps://www.optimizely.com/sample-size-calculator/\\n- Business-friendly language\\n- Duration estimates\\n\\n**AB Test Guide Calculator**\\nhttps://www.abtestguide.com/calc/\\n- Includes Bayesian option\\n- Multiple test types\\n\\n**VWO Duration Calculator**\\nhttps://vwo.com/tools/ab-test-duration-calculator/\\n- Duration-focused\\n- Good for planning\\n\\n---\\n\\n## Adjusting for Multiple Variants\\n\\nWith more than 2 variants (A/B/n tests), you need more sample:\\n\\n| Variants | Multiplier |\\n|----------|------------|\\n| 2 (A/B) | 1x |\\n| 3 (A/B/C) | ~1.5x |\\n| 4 (A/B/C/D) | ~2x |\\n| 5+ | Consider reducing variants |\\n\\n**Why?** More comparisons increase chance of false positives. You're comparing:\\n- A vs B\\n- A vs C\\n- B vs C (sometimes)\\n\\nApply Bonferroni correction or use tools that handle this automatically.\\n\\n---\\n\\n## Common Sample Size Mistakes\\n\\n### 1. Underpowered tests\\n**Problem**: Not enough sample to detect realistic effects\\n**Fix**: Be realistic about MDE, get more traffic, or don't test\\n\\n### 2. Overpowered tests\\n**Problem**: Waiting for sample size when you already have significance\\n**Fix**: This is actually fine—you committed to sample size, honor it\\n\\n### 3. Wrong baseline rate\\n**Problem**: Using wrong conversion rate for calculation\\n**Fix**: Use the specific metric and page, not site-wide averages\\n\\n### 4. Ignoring segments\\n**Problem**: Calculating for full traffic, then analyzing segments\\n**Fix**: If you plan segment analysis, calculate sample for smallest segment\\n\\n### 5. Testing too many things\\n**Problem**: Dividing traffic too many ways\\n**Fix**: Prioritize ruthlessly, run fewer concurrent tests\\n\\n---\\n\\n## When Sample Size Requirements Are Too High\\n\\nOptions when you can't get enough traffic:\\n\\n1. **Increase MDE**: Accept only detecting larger effects (20%+ lift)\\n2. **Lower confidence**: Use 90% instead of 95% (risky, document it)\\n3. **Reduce variants**: Test only the most promising variant\\n4. **Combine traffic**: Test across multiple similar pages\\n5. **Test upstream**: Test earlier in funnel where traffic is higher\\n6. **Don't test**: Make decision based on qualitative data instead\\n7. **Longer test**: Accept longer duration (weeks/months)\\n\\n---\\n\\n## Sequential Testing\\n\\nIf you must check results before reaching sample size:\\n\\n### What is it?\\nStatistical method that adjusts for multiple looks at data.\\n\\n### When to use\\n- High-risk changes\\n- Need to stop bad variants early\\n- Time-sensitive decisions\\n\\n### Tools that support it\\n- Optimizely (Stats Accelerator)\\n- VWO (SmartStats)\\n- PostHog (Bayesian approach)\\n\\n### Tradeoff\\n- More flexibility to stop early\\n- Slightly larger sample size requirement\\n- More complex analysis\\n\\n---\\n\\n## Quick Decision Framework\\n\\n### Can I run this test?\\n\\n```\\nDaily traffic to page: _____\\nBaseline conversion rate: _____\\nMDE I care about: _____\\n\\nSample needed per variant: _____ (from tables above)\\nDays to run: Sample / Daily traffic = _____\\n\\nIf days > 60: Consider alternatives\\nIf days > 30: Acceptable for high-impact tests\\nIf days < 14: Likely feasible\\nIf days < 7: Easy to run, consider running longer anyway\\n```\\n\"},\n {\"skill\":\"ab-testing\",\"ref\":\"test-templates\",\"body\":\"# A/B Test Templates Reference\\n\\nTemplates for planning, documenting, and analyzing experiments.\\n\\n## Contents\\n- Test Plan Template\\n- Results Documentation Template\\n- Test Repository Entry Template\\n- Quick Test Brief Template\\n- Stakeholder Update Template\\n- Experiment Prioritization Scorecard\\n- Hypothesis Bank Template\\n\\n## Test Plan Template\\n\\n```markdown\\n# A/B Test: [Name]\\n\\n## Overview\\n- **Owner**: [Name]\\n- **Test ID**: [ID in testing tool]\\n- **Page/Feature**: [What's being tested]\\n- **Planned dates**: [Start] - [End]\\n\\n## Hypothesis\\n\\nBecause [observation/data],\\nwe believe [change]\\nwill cause [expected outcome]\\nfor [audience].\\nWe'll know this is true when [metrics].\\n\\n## Test Design\\n\\n| Element | Details |\\n|---------|---------|\\n| Test type | A/B / A/B/n / MVT |\\n| Duration | X weeks |\\n| Sample size | X per variant |\\n| Traffic allocation | 50/50 |\\n| Tool | [Tool name] |\\n| Implementation | Client-side / Server-side |\\n\\n## Variants\\n\\n### Control (A)\\n[Screenshot]\\n- Current experience\\n- [Key details about current state]\\n\\n### Variant (B)\\n[Screenshot or mockup]\\n- [Specific change #1]\\n- [Specific change #2]\\n- Rationale: [Why we think this will win]\\n\\n## Metrics\\n\\n### Primary\\n- **Metric**: [metric name]\\n- **Definition**: [how it's calculated]\\n- **Current baseline**: [X%]\\n- **Minimum detectable effect**: [X%]\\n\\n### Secondary\\n- [Metric 1]: [what it tells us]\\n- [Metric 2]: [what it tells us]\\n- [Metric 3]: [what it tells us]\\n\\n### Guardrails\\n- [Metric that shouldn't get worse]\\n- [Another safety metric]\\n\\n## Segment Analysis Plan\\n- Mobile vs. desktop\\n- New vs. returning visitors\\n- Traffic source\\n- [Other relevant segments]\\n\\n## Success Criteria\\n- Winner: [Primary metric improves by X% with 95% confidence]\\n- Loser: [Primary metric decreases significantly]\\n- Inconclusive: [What we'll do if no significant result]\\n\\n## Pre-Launch Checklist\\n- [ ] Hypothesis documented and reviewed\\n- [ ] Primary metric defined and trackable\\n- [ ] Sample size calculated\\n- [ ] Test duration estimated\\n- [ ] Variants implemented correctly\\n- [ ] Tracking verified in all variants\\n- [ ] QA completed on all variants\\n- [ ] Stakeholders informed\\n- [ ] Calendar hold for analysis date\\n```\\n\\n---\\n\\n## Results Documentation Template\\n\\n```markdown\\n# A/B Test Results: [Name]\\n\\n## Summary\\n| Element | Value |\\n|---------|-------|\\n| Test ID | [ID] |\\n| Dates | [Start] - [End] |\\n| Duration | X days |\\n| Result | Winner / Loser / Inconclusive |\\n| Decision | [What we're doing] |\\n\\n## Hypothesis (Reminder)\\n[Copy from test plan]\\n\\n## Results\\n\\n### Sample Size\\n| Variant | Target | Actual | % of target |\\n|---------|--------|--------|-------------|\\n| Control | X | Y | Z% |\\n| Variant | X | Y | Z% |\\n\\n### Primary Metric: [Metric Name]\\n| Variant | Value | 95% CI | vs. Control |\\n|---------|-------|--------|-------------|\\n| Control | X% | [X%, Y%] | — |\\n| Variant | X% | [X%, Y%] | +X% |\\n\\n**Statistical significance**: p = X.XX (95% = sig / not sig)\\n**Practical significance**: [Is this lift meaningful for the business?]\\n\\n### Secondary Metrics\\n\\n| Metric | Control | Variant | Change | Significant? |\\n|--------|---------|---------|--------|--------------|\\n| [Metric 1] | X | Y | +Z% | Yes/No |\\n| [Metric 2] | X | Y | +Z% | Yes/No |\\n\\n### Guardrail Metrics\\n\\n| Metric | Control | Variant | Change | Concern? |\\n|--------|---------|---------|--------|----------|\\n| [Metric 1] | X | Y | +Z% | Yes/No |\\n\\n### Segment Analysis\\n\\n**Mobile vs. Desktop**\\n| Segment | Control | Variant | Lift |\\n|---------|---------|---------|------|\\n| Mobile | X% | Y% | +Z% |\\n| Desktop | X% | Y% | +Z% |\\n\\n**New vs. Returning**\\n| Segment | Control | Variant | Lift |\\n|---------|---------|---------|------|\\n| New | X% | Y% | +Z% |\\n| Returning | X% | Y% | +Z% |\\n\\n## Interpretation\\n\\n### What happened?\\n[Explanation of results in plain language]\\n\\n### Why do we think this happened?\\n[Analysis and reasoning]\\n\\n### Caveats\\n[Any limitations, external factors, or concerns]\\n\\n## Decision\\n\\n**Winner**: [Control / Variant]\\n\\n**Action**: [Implement variant / Keep control / Re-test]\\n\\n**Timeline**: [When changes will be implemented]\\n\\n## Learnings\\n\\n### What we learned\\n- [Key insight 1]\\n- [Key insight 2]\\n\\n### What to test next\\n- [Follow-up test idea 1]\\n- [Follow-up test idea 2]\\n\\n### Impact\\n- **Projected lift**: [X% improvement in Y metric]\\n- **Business impact**: [Revenue, conversions, etc.]\\n```\\n\\n---\\n\\n## Test Repository Entry Template\\n\\nFor tracking all tests in a central location:\\n\\n```markdown\\n| Test ID | Name | Page | Dates | Primary Metric | Result | Lift | Link |\\n|---------|------|------|-------|----------------|--------|------|------|\\n| 001 | Hero headline test | Homepage | 1/1-1/15 | CTR | Winner | +12% | [Link] |\\n| 002 | Pricing table layout | Pricing | 1/10-1/31 | Plan selection | Loser | -5% | [Link] |\\n| 003 | Signup form fields | Signup | 2/1-2/14 | Completion | Inconclusive | +2% | [Link] |\\n```\\n\\n---\\n\\n## Quick Test Brief Template\\n\\nFor simple tests that don't need full documentation:\\n\\n```markdown\\n## [Test Name]\\n\\n**What**: [One sentence description]\\n**Why**: [One sentence hypothesis]\\n**Metric**: [Primary metric]\\n**Duration**: [X weeks]\\n**Result**: [TBD / Winner / Loser / Inconclusive]\\n**Learnings**: [Key takeaway]\\n```\\n\\n---\\n\\n## Stakeholder Update Template\\n\\n```markdown\\n## A/B Test Update: [Name]\\n\\n**Status**: Running / Complete\\n**Days remaining**: X (or complete)\\n**Current sample**: X% of target\\n\\n### Preliminary observations\\n[What we're seeing - without making decisions yet]\\n\\n### Next steps\\n[What happens next]\\n\\n### Timeline\\n- [Date]: Analysis complete\\n- [Date]: Decision and recommendation\\n- [Date]: Implementation (if winner)\\n```\\n\\n---\\n\\n## Experiment Prioritization Scorecard\\n\\nFor deciding which tests to run:\\n\\n| Factor | Weight | Test A | Test B | Test C |\\n|--------|--------|--------|--------|--------|\\n| Potential impact | 30% | | | |\\n| Confidence in hypothesis | 25% | | | |\\n| Ease of implementation | 20% | | | |\\n| Risk if wrong | 15% | | | |\\n| Strategic alignment | 10% | | | |\\n| **Total** | | | | |\\n\\nScoring: 1-5 (5 = best)\\n\\n---\\n\\n## Hypothesis Bank Template\\n\\nFor collecting test ideas:\\n\\n```markdown\\n| ID | Page/Area | Observation | Hypothesis | Potential Impact | Status |\\n|----|-----------|-------------|------------|------------------|--------|\\n| H1 | Homepage | Low scroll depth | Shorter hero will increase scroll | High | Testing |\\n| H2 | Pricing | Users compare plans | Comparison table will help | Medium | Backlog |\\n| H3 | Signup | Drop-off at email | Social login will increase completion | Medium | Backlog |\\n```\\n\"},\n {\"skill\":\"ad-creative\",\"ref\":\"generative-tools\",\"body\":\"# Generative AI Tools for Ad Creative\\n\\nReference for using AI image generators, video generators, and code-based video tools to produce ad visuals at scale.\\n\\n---\\n\\n## When to Use Generative Tools\\n\\n| Need | Tool Category | Best Fit |\\n|------|---------------|----------|\\n| Static ad images (banners, social) | Image generation | ChatGPT Images 2.0, Nano Banana Pro, Flux, Ideogram |\\n| Ad images with text overlays | Image generation (text-capable) | Ideogram, Nano Banana Pro |\\n| Short video ads (6-30 sec) | Video generation | Veo, Kling, Runway, Sora, Seedance |\\n| Video ads with voiceover | Video gen + voice | Veo/Sora (native), or Runway + ElevenLabs |\\n| Voiceover tracks for ads | Voice generation | ElevenLabs, OpenAI TTS, Cartesia |\\n| Multi-language ad versions | Voice generation | ElevenLabs, PlayHT |\\n| Brand voice cloning | Voice generation | ElevenLabs, Resemble AI |\\n| Product mockups and variations | Image generation + references | Flux (multi-image reference) |\\n| Templated video ads at scale | Code-based video | Remotion |\\n| Personalized video (name, data) | Code-based video | Remotion |\\n| Brand-consistent variations | Image gen + style refs | Flux, Ideogram, Nano Banana Pro |\\n\\n---\\n\\n## Image Generation\\n\\n### Nano Banana Pro (Gemini)\\n\\nGoogle DeepMind's image generation model, available through the Gemini API.\\n\\n**Best for:** High-quality ad images, product visuals, text rendering\\n**API:** Gemini API (Google AI Studio, Vertex AI)\\n**Pricing:** ~$0.04/image (Gemini 2.5 Flash Image), ~$0.24/4K image (Nano Banana Pro)\\n\\n**Strengths:**\\n- Strong text rendering in images (logos, headlines)\\n- Native image editing (modify existing images with prompts)\\n- Available through the same Gemini API used for text generation\\n- Supports both generation and editing in one model\\n\\n**Ad creative use cases:**\\n- Generate social media ad images from text descriptions\\n- Create product mockup variations\\n- Edit existing ad images (swap backgrounds, change colors)\\n- Generate images with headline text baked in\\n\\n**API example:**\\n```bash\\n# Using the Gemini API for image generation\\ncurl -X POST \\\"https://generativelanguage.googleapis.com/v1beta/models/gemini-2.5-flash-image:generateContent\\\" \\\\\\n -H \\\"Content-Type: application/json\\\" \\\\\\n -H \\\"x-goog-api-key: $GEMINI_API_KEY\\\" \\\\\\n -d '{\\n \\\"contents\\\": [{\\\"parts\\\": [{\\\"text\\\": \\\"Create a clean, modern social media ad image for a project management tool. Show a laptop with a kanban board interface. Bright, professional, 16:9 ratio.\\\"}]}],\\n \\\"generationConfig\\\": {\\\"responseModalities\\\": [\\\"TEXT\\\", \\\"IMAGE\\\"]}\\n }'\\n```\\n\\n**Docs:** [Gemini Image Generation](https://ai.google.dev/gemini-api/docs/image-generation)\\n\\n---\\n\\n### Flux (Black Forest Labs)\\n\\nOpen-weight image generation models with API access through Replicate and BFL's native API.\\n\\n**Best for:** Photorealistic images, brand-consistent variations, multi-reference generation\\n**API:** Replicate, BFL API, fal.ai\\n**Pricing:** ~$0.01-0.06/image depending on model and resolution\\n\\n**Model variants:**\\n| Model | Speed | Quality | Cost | Best For |\\n|-------|-------|---------|------|----------|\\n| Flux 2 Pro | ~6 sec | Highest | $0.015/MP | Final production assets |\\n| Flux 2 Flex | ~22 sec | High + editing | $0.06/MP | Iterative editing |\\n| Flux 2 Dev | ~2.5 sec | Good | $0.012/MP | Rapid prototyping |\\n| Flux 2 Klein | Fastest | Good | Lowest | High-volume batch generation |\\n\\n**Strengths:**\\n- Multi-image reference (up to 8 images) for consistent identity across ads\\n- Product consistency — same product in different contexts\\n- Style transfer from reference images\\n- Open-weight Dev model for self-hosting\\n\\n**Ad creative use cases:**\\n- Generate 50+ ad variations with consistent product/person identity\\n- Create product-in-context images (your SaaS on different devices)\\n- Style-match to existing brand assets using reference images\\n- Rapid A/B test image variations\\n\\n**Docs:** [Replicate Flux](https://replicate.com/black-forest-labs/flux-2-pro), [BFL API](https://docs.bfl.ml/)\\n\\n---\\n\\n### Ideogram\\n\\nSpecialized in typography and text rendering within images.\\n\\n**Best for:** Ad banners with text, branded graphics, social ad images with headlines\\n**API:** Ideogram API, Runware\\n**Pricing:** ~$0.06/image (API), ~$0.009/image (subscription)\\n\\n**Strengths:**\\n- Best-in-class text rendering (~90% accuracy vs ~30% for most tools)\\n- Style reference system (upload up to 3 reference images)\\n- 4.3 billion style presets for consistent brand aesthetics\\n- Strong at logos and branded typography\\n\\n**Ad creative use cases:**\\n- Generate ad banners with headline text directly in the image\\n- Create social media graphics with branded text overlays\\n- Produce multiple design variations with consistent typography\\n- Generate promotional materials without needing a designer for each iteration\\n\\n**Docs:** [Ideogram API](https://developer.ideogram.ai/), [Ideogram](https://ideogram.ai/)\\n\\n---\\n\\n### Other Image Tools\\n\\n| Tool | Best For | API Status | Notes |\\n|------|----------|------------|-------|\\n| **DALL-E 3** (OpenAI) | General image generation | Official API | Integrated with ChatGPT, good text rendering |\\n| **Midjourney** | Artistic, high-aesthetic images | No official public API | Discord-based; unofficial APIs exist but risk bans |\\n| **Stable Diffusion** | Self-hosted, customizable | Open source | Best for teams with GPU infrastructure |\\n\\n---\\n\\n## Video Generation\\n\\n### Google Veo\\n\\nGoogle DeepMind's video generation model, available through the Gemini API and Vertex AI.\\n\\n**Best for:** High-quality video ads with native audio, vertical video for social\\n**API:** Gemini API, Vertex AI\\n**Pricing:** ~$0.15/sec (Veo 3.1 Fast), ~$0.40/sec (Veo 3.1 Standard)\\n\\n**Capabilities:**\\n- Up to 60 seconds at 1080p\\n- Native audio generation (dialogue, sound effects, ambient)\\n- Vertical 9:16 output for Stories/Reels/Shorts\\n- Upscale to 4K\\n- Text-to-video and image-to-video\\n\\n**Ad creative use cases:**\\n- Generate short video ads (15-30 sec) from text descriptions\\n- Create vertical video ads for TikTok, Reels, Shorts\\n- Produce product demos with voiceover\\n- Generate multiple video variations from the same prompt with different styles\\n\\n**Docs:** [Veo on Vertex AI](https://cloud.google.com/vertex-ai/generative-ai/docs/video/overview)\\n\\n---\\n\\n### Kling (Kuaishou)\\n\\nVideo generation with simultaneous audio-visual generation and camera controls.\\n\\n**Best for:** Cinematic video ads, longer-form content, audio-synced video\\n**API:** Kling API, PiAPI, fal.ai\\n**Pricing:** ~$0.09/sec (via fal.ai third-party)\\n\\n**Capabilities:**\\n- Up to 3 minutes at 1080p/30-48fps\\n- Simultaneous audio-visual generation (Kling 2.6)\\n- Text-to-video and image-to-video\\n- Motion and camera controls\\n\\n**Ad creative use cases:**\\n- Longer product explainer videos\\n- Cinematic brand videos with synchronized audio\\n- Animate product images into video ads\\n\\n**Docs:** [Kling AI Developer](https://klingai.com/global/dev/model/video)\\n\\n---\\n\\n### Runway\\n\\nVideo generation and editing platform with strong controllability.\\n\\n**Best for:** Controlled video generation, style-consistent content, editing existing footage\\n**API:** Runway Developer Portal\\n\\n**Capabilities:**\\n- Gen-4: Character/scene consistency across shots\\n- Motion brush and camera controls\\n- Image-to-video with reference images\\n- Video-to-video style transfer\\n\\n**Ad creative use cases:**\\n- Generate video ads with consistent characters/products across scenes\\n- Style-transfer existing footage to match brand aesthetics\\n- Extend or remix existing video content\\n\\n**Docs:** [Runway API](https://docs.dev.runwayml.com/)\\n\\n---\\n\\n### Sora 2 (OpenAI)\\n\\nOpenAI's video generation model with synchronized audio.\\n\\n**Best for:** High-fidelity video with dialogue and sound\\n**API:** OpenAI API\\n**Pricing:** Free tier available; Pro from $0.10-0.50/sec depending on resolution\\n\\n**Capabilities:**\\n- Up to 60 seconds with synchronized audio\\n- Dialogue, sound effects, and ambient audio\\n- sora-2 (fast) and sora-2-pro (quality) variants\\n- Text-to-video and image-to-video\\n\\n**Ad creative use cases:**\\n- Video testimonials and talking-head style ads\\n- Product demo videos with narration\\n- Narrative brand videos\\n\\n**Docs:** [OpenAI Video Generation](https://platform.openai.com/docs/guides/video-generation)\\n\\n---\\n\\n### Seedance 2.0 (ByteDance)\\n\\nByteDance's video generation model with simultaneous audio-visual generation and multimodal inputs.\\n\\n**Best for:** Fast, affordable video ads with native audio, multimodal reference inputs\\n**API:** BytePlus (official), Replicate, WaveSpeedAI, fal.ai (third-party); OpenAI-compatible API format\\n**Pricing:** ~$0.10-0.80/min depending on resolution (estimated 10-100x cheaper than Sora 2 per clip)\\n\\n**Capabilities:**\\n- Up to 20 seconds at up to 2K resolution\\n- Simultaneous audio-visual generation (Dual-Branch Diffusion Transformer)\\n- Text-to-video and image-to-video\\n- Up to 12 reference files for multimodal input\\n- OpenAI-compatible API structure\\n\\n**Ad creative use cases:**\\n- High-volume short video ad production at low cost\\n- Video ads with synchronized voiceover and sound effects in one pass\\n- Multi-reference generation (feed product images, brand assets, style references)\\n- Rapid iteration on video ad concepts\\n\\n**Docs:** [Seedance](https://seed.bytedance.com/en/seedance2_0)\\n\\n---\\n\\n### Higgsfield\\n\\nFull-stack video creation platform with cinematic camera controls.\\n\\n**Best for:** Social video ads, cinematic style, mobile-first content\\n**Platform:** [higgsfield.ai](https://higgsfield.ai/)\\n\\n**Capabilities:**\\n- 50+ professional camera movements (zooms, pans, FPV drone shots)\\n- Image-to-video animation\\n- Built-in editing, transitions, and keyframing\\n- All-in-one workflow: image gen, animation, editing\\n\\n**Ad creative use cases:**\\n- Social media video ads with cinematic feel\\n- Animate product images into dynamic video\\n- Create multiple video variations with different camera styles\\n- Quick-turn video content for social campaigns\\n\\n---\\n\\n### Video Tool Comparison\\n\\n| Tool | Max Length | Audio | Resolution | API | Best For |\\n|------|-----------|-------|------------|-----|----------|\\n| **Veo 3.1** | 60 sec | Native | 1080p/4K | Gemini | Vertical social video |\\n| **Kling 2.6** | 3 min | Native | 1080p | Third-party | Longer cinematic |\\n| **Runway Gen-4** | 10 sec | No | 1080p | Official | Controlled, consistent |\\n| **Sora 2** | 60 sec | Native | 1080p | Official | Dialogue-heavy |\\n| **Seedance 2.0** | 20 sec | Native | 2K | Official + third-party | Affordable high-volume |\\n| **Higgsfield** | Varies | Yes | 1080p | Web-based | Social, mobile-first |\\n\\n---\\n\\n## Voice & Audio Generation\\n\\nFor layering realistic voiceovers onto video ads, adding narration to product demos, or generating audio for Remotion-rendered videos. These tools turn ad scripts into natural-sounding voice tracks.\\n\\n### When to Use Voice Tools\\n\\nMany video generators (Veo, Kling, Sora, Seedance) now include native audio. Use standalone voice tools when you need:\\n\\n- **Voiceover on silent video** — Runway Gen-4 and Remotion produce silent output\\n- **Brand voice consistency** — Clone a specific voice for all ads\\n- **Multi-language versions** — Same ad script in 20+ languages\\n- **Script iteration** — Re-record voiceover without reshooting video\\n- **Precise control** — Exact timing, emotion, and pacing\\n\\n---\\n\\n### ElevenLabs\\n\\nThe market leader in realistic voice generation and voice cloning.\\n\\n**Best for:** Most natural-sounding voiceovers, brand voice cloning, multilingual\\n**API:** REST API with streaming support\\n**Pricing:** ~$0.12-0.30 per 1,000 characters depending on plan; starts at $5/month\\n\\n**Capabilities:**\\n- 29+ languages with natural accent and intonation\\n- Voice cloning from short audio clips (instant) or longer recordings (professional)\\n- Emotion and style control\\n- Streaming for real-time generation\\n- Voice library with hundreds of pre-built voices\\n\\n**Ad creative use cases:**\\n- Generate voiceover tracks for video ads\\n- Clone your brand spokesperson's voice for all ad variations\\n- Produce the same ad in 10+ languages from one script\\n- A/B test different voice styles (authoritative vs. friendly vs. urgent)\\n\\n**API example:**\\n```bash\\ncurl -X POST \\\"https://api.elevenlabs.io/v1/text-to-speech/{voice_id}\\\" \\\\\\n -H \\\"xi-api-key: $ELEVENLABS_API_KEY\\\" \\\\\\n -H \\\"Content-Type: application/json\\\" \\\\\\n -d '{\\n \\\"text\\\": \\\"Stop wasting hours on manual reporting. Try DataFlow free for 14 days.\\\",\\n \\\"model_id\\\": \\\"eleven_multilingual_v2\\\",\\n \\\"voice_settings\\\": {\\\"stability\\\": 0.5, \\\"similarity_boost\\\": 0.75}\\n }' --output voiceover.mp3\\n```\\n\\n**Docs:** [ElevenLabs API](https://elevenlabs.io/docs/api-reference/text-to-speech)\\n\\n---\\n\\n### OpenAI TTS\\n\\nSimple, affordable text-to-speech built into the OpenAI API.\\n\\n**Best for:** Quick voiceovers, cost-effective at scale, simple integration\\n**API:** OpenAI API (same SDK as GPT/DALL-E)\\n**Pricing:** $15/million chars (standard), $30/million chars (HD); ~$0.015/min with gpt-4o-mini-tts\\n\\n**Capabilities:**\\n- 13 built-in voices (no custom cloning)\\n- Multiple languages\\n- Real-time streaming\\n- HD quality option\\n- Simple API — same SDK you already use for GPT\\n\\n**Ad creative use cases:**\\n- Fast, cheap voiceover for draft/test ad versions\\n- High-volume narration at low cost\\n- Prototype ad audio before investing in premium voice\\n\\n**Docs:** [OpenAI TTS](https://platform.openai.com/docs/guides/text-to-speech)\\n\\n---\\n\\n### Cartesia Sonic\\n\\nUltra-low latency voice generation built for real-time applications.\\n\\n**Best for:** Real-time voice, lowest latency, emotional expressiveness\\n**API:** REST + WebSocket streaming\\n**Pricing:** Starts at $5/month; pay-as-you-go from $0.03/min\\n\\n**Capabilities:**\\n- 40ms time-to-first-audio (fastest in class)\\n- 15+ languages\\n- Nonverbal expressiveness: laughter, breathing, emotional inflections\\n- Sonic Turbo for even lower latency\\n- Streaming API for real-time generation\\n\\n**Ad creative use cases:**\\n- Real-time ad preview during creative iteration\\n- Interactive demo videos with dynamic narration\\n- Ads requiring natural laughter, sighs, or emotional reactions\\n\\n**Docs:** [Cartesia Sonic](https://docs.cartesia.ai/build-with-cartesia/tts-models/latest)\\n\\n---\\n\\n### Voicebox (Open Source)\\n\\nFree, local-first voice synthesis studio powered by Qwen3-TTS. The open-source alternative to ElevenLabs.\\n\\n**Best for:** Free voice cloning, local/private generation, zero-cost batch production\\n**API:** Local REST API at `http://localhost:8000`\\n**Pricing:** Free (MIT license). Runs entirely on your machine.\\n**Stack:** Tauri (Rust) + React + FastAPI (Python)\\n\\n**Capabilities:**\\n- Voice cloning from short audio samples via Qwen3-TTS\\n- Multi-language support (English, Chinese, more planned)\\n- Multi-track timeline editor for composing conversations\\n- 4-5x faster inference on Apple Silicon via MLX Metal acceleration\\n- Local REST API for programmatic generation\\n- No cloud dependency — all processing on-device\\n\\n**Ad creative use cases:**\\n- Free voice cloning for brand spokesperson across all ad variations\\n- Batch generate voiceovers without per-character costs\\n- Private/local generation when ad content is sensitive or pre-launch\\n- Prototype voice variations before committing to a paid service\\n\\n**API example:**\\n```bash\\ncurl -X POST http://localhost:8000/generate \\\\\\n -H \\\"Content-Type: application/json\\\" \\\\\\n -d '{\\\"text\\\": \\\"Stop wasting hours on manual reporting.\\\", \\\"profile_id\\\": \\\"abc123\\\", \\\"language\\\": \\\"en\\\"}'\\n```\\n\\n**Install:** Desktop apps for macOS and Windows at [voicebox.sh](https://voicebox.sh), or build from source:\\n```bash\\ngit clone https://github.com/jamiepine/voicebox.git\\ncd voicebox && make setup && make dev\\n```\\n\\n**Docs:** [GitHub](https://github.com/jamiepine/voicebox)\\n\\n---\\n\\n### Other Voice Tools\\n\\n| Tool | Best For | Differentiator | API |\\n|------|----------|---------------|-----|\\n| **PlayHT** | Large voice library, low latency | 900+ voices, <300ms latency, ultra-realistic | [play.ht](https://play.ht/) |\\n| **Resemble AI** | Enterprise voice cloning | On-premise deployment, real-time speech-to-speech | [resemble.ai](https://www.resemble.ai/) |\\n| **WellSaid Labs** | Ethical, commercial-safe voices | Voices from compensated actors, safe for commercial use | [wellsaid.io](https://www.wellsaid.io/) |\\n| **Fish Audio** | Budget-friendly, emotion control | ~50-70% cheaper than ElevenLabs, emotion tags | [fish.audio](https://fish.audio/) |\\n| **Murf AI** | Non-technical teams | Browser-based studio, 200+ voices | [murf.ai](https://murf.ai/) |\\n| **Google Cloud TTS** | Google ecosystem, scale | 220+ voices, 40+ languages, enterprise SLAs | [Google TTS](https://cloud.google.com/text-to-speech) |\\n| **Amazon Polly** | AWS ecosystem, cost | Neural voices, SSML control, cheap at volume | [Amazon Polly](https://aws.amazon.com/polly/) |\\n\\n---\\n\\n### Voice Tool Comparison\\n\\n| Tool | Quality | Cloning | Languages | Latency | Price/1K chars |\\n|------|---------|---------|-----------|---------|----------------|\\n| **ElevenLabs** | Best | Yes (instant + pro) | 29+ | ~200ms | $0.12-0.30 |\\n| **OpenAI TTS** | Good | No | 13+ | ~300ms | $0.015-0.030 |\\n| **Cartesia Sonic** | Very good | No | 15+ | ~40ms | ~$0.03/min |\\n| **PlayHT** | Very good | Yes | 140+ | <300ms | ~$0.10-0.20 |\\n| **Fish Audio** | Good | Yes | 13+ | ~200ms | ~$0.05-0.10 |\\n| **WellSaid** | Very good | No (actor voices) | English | ~300ms | Custom pricing |\\n| **Voicebox** | Good | Yes (local) | 2+ | Local | Free (open source) |\\n\\n### Choosing a Voice Tool\\n\\n```\\nNeed voiceover for ads?\\n├── Need to clone a specific brand voice?\\n│ ├── Best quality → ElevenLabs\\n│ ├── Enterprise/on-premise → Resemble AI\\n│ └── Budget-friendly → Fish Audio, PlayHT\\n├── Need multilingual (same ad, many languages)?\\n│ ├── Most languages → PlayHT (140+)\\n│ └── Best quality → ElevenLabs (29+)\\n├── Need free / open source / local?\\n│ └── Voicebox (MIT, runs on your machine)\\n├── Need cheap, fast, good-enough?\\n│ └── OpenAI TTS ($0.015/min)\\n├── Need commercially-safe licensing?\\n│ └── WellSaid Labs (actor-compensated voices)\\n└── Need real-time/interactive?\\n └── Cartesia Sonic (40ms TTFA)\\n```\\n\\n### Workflow: Voice + Video\\n\\n```\\n1. Write ad script (use ad-creative skill for copy)\\n2. Generate voiceover with ElevenLabs/OpenAI TTS\\n3. Generate or render video:\\n a. Silent video from Runway/Remotion → layer voice track\\n b. Or use Veo/Sora/Seedance with native audio (skip separate VO)\\n4. Combine with ffmpeg if layering separately:\\n ffmpeg -i video.mp4 -i voiceover.mp3 -c:v copy -c:a aac output.mp4\\n5. Generate variations (different scripts, voices, or languages)\\n```\\n\\n---\\n\\n## Code-Based Video: Remotion\\n\\nFor templated, data-driven video ads at scale, Remotion is the best option. Unlike AI video generators that produce unique video from prompts, Remotion uses React code to render deterministic, brand-perfect video from templates and data.\\n\\n**Best for:** Templated ad variations, personalized video, brand-consistent production\\n**Stack:** React + TypeScript\\n**Pricing:** Free for individuals/small teams; commercial license required for 4+ employees\\n**Docs:** [remotion.dev](https://www.remotion.dev/)\\n\\n### Why Remotion for Ads\\n\\n| AI Video Generators | Remotion |\\n|---------------------|----------|\\n| Unique output each time | Deterministic, pixel-perfect |\\n| Prompt-based, less control | Full code control over every frame |\\n| Hard to match brand exactly | Exact brand colors, fonts, spacing |\\n| One-at-a-time generation | Batch render hundreds from data |\\n| No dynamic data insertion | Personalize with names, prices, stats |\\n\\n### Ad Creative Use Cases\\n\\n**1. Dynamic product ads**\\nFeed a JSON array of products and render a unique video ad for each:\\n```tsx\\n// Simplified Remotion component for product ads\\nexport const ProductAd: React.FC<{\\n productName: string;\\n price: string;\\n imageUrl: string;\\n tagline: string;\\n}> = ({productName, price, imageUrl, tagline}) => {\\n return (\\n \\n

\\n

{productName}

\\n

{tagline}

\\n

{price}

\\n

Shop Now

\\n \\n );\\n};\\n```\\n\\n**2. A/B test video variations**\\nRender the same template with different headlines, CTAs, or color schemes:\\n```tsx\\nconst variations = [\\n {headline: \\\"Save 50% Today\\\", cta: \\\"Get the Deal\\\", theme: \\\"urgent\\\"},\\n {headline: \\\"Join 10K+ Teams\\\", cta: \\\"Start Free\\\", theme: \\\"social-proof\\\"},\\n {headline: \\\"Built for Speed\\\", cta: \\\"Try It Now\\\", theme: \\\"benefit\\\"},\\n];\\n// Render all variations programmatically\\n```\\n\\n**3. Personalized outreach videos**\\nGenerate videos addressing prospects by name for cold outreach or sales.\\n\\n**4. Social ad batch production**\\nRender the same content across different aspect ratios:\\n- 1:1 for feed\\n- 9:16 for Stories/Reels\\n- 16:9 for YouTube\\n\\n### Remotion Workflow for Ad Creative\\n\\n```\\n1. Design template in React (or use AI to generate the component)\\n2. Define data schema (products, headlines, CTAs, images)\\n3. Feed data array into template\\n4. Batch render all variations\\n5. Upload to ad platform\\n```\\n\\n### Getting Started\\n\\n```bash\\n# Create a new Remotion project\\nnpx create-video@latest\\n\\n# Render a single video\\nnpx remotion render src/index.ts MyComposition out/video.mp4\\n\\n# Batch render from data\\nnpx remotion render src/index.ts MyComposition --props='{\\\"data\\\": [...]}'\\n```\\n\\n---\\n\\n## Choosing the Right Tool\\n\\n### Decision Tree\\n\\n```\\nNeed video ads?\\n├── Templated, data-driven (same structure, different data)\\n│ └── Use Remotion\\n├── Unique creative from prompts (exploratory)\\n│ ├── Need dialogue/voiceover? → Sora 2, Veo 3.1, Kling 2.6, Seedance 2.0\\n│ ├── Need consistency across scenes? → Runway Gen-4\\n│ ├── Need vertical social video? → Veo 3.1 (native 9:16)\\n│ ├── Need high volume at low cost? → Seedance 2.0\\n│ └── Need cinematic camera work? → Higgsfield, Kling\\n└── Both → Use AI gen for hero creative, Remotion for variations\\n\\nNeed image ads?\\n├── Need text/headlines in image? → Ideogram\\n├── Need product consistency across variations? → Flux (multi-ref)\\n├── Need quick iterations on existing images? → Nano Banana Pro\\n├── Need highest visual quality? → Flux Pro, Midjourney\\n└── Need high volume at low cost? → Flux Klein, Nano Banana\\n```\\n\\n### Cost Comparison for 100 Ad Variations\\n\\n| Approach | Tool | Approximate Cost |\\n|----------|------|-----------------|\\n| 100 static images | Nano Banana Pro | ~$4-24 |\\n| 100 static images | Flux Dev | ~$1-2 |\\n| 100 static images | Ideogram API | ~$6 |\\n| 100 × 15-sec videos | Veo 3.1 Fast | ~$225 |\\n| 100 × 15-sec videos | Remotion (templated) | ~$0 (self-hosted render) |\\n| 10 hero videos + 90 templated | Veo + Remotion | ~$22 + render time |\\n\\n### Recommended Workflow for Scaled Ad Production\\n\\n1. **Generate hero creative** with AI (Nano Banana, Flux, Veo) — high-quality, exploratory\\n2. **Build templates** in Remotion based on winning creative patterns\\n3. **Batch produce variations** with Remotion using data (products, headlines, CTAs)\\n4. **Iterate** — use AI tools for new angles, Remotion for scale\\n\\nThis hybrid approach gives you the creative exploration of AI generators and the consistency and scale of code-based rendering.\\n\\n---\\n\\n## Platform-Specific Image Specs\\n\\nWhen generating images for ads, request the correct dimensions:\\n\\n| Platform | Placement | Aspect Ratio | Recommended Size |\\n|----------|-----------|-------------|-----------------|\\n| Meta Feed | Single image | 1:1 | 1080x1080 |\\n| Meta Stories/Reels | Vertical | 9:16 | 1080x1920 |\\n| Meta Carousel | Square | 1:1 | 1080x1080 |\\n| Google Display | Landscape | 1.91:1 | 1200x628 |\\n| Google Display | Square | 1:1 | 1200x1200 |\\n| LinkedIn Feed | Landscape | 1.91:1 | 1200x627 |\\n| LinkedIn Feed | Square | 1:1 | 1200x1200 |\\n| TikTok Feed | Vertical | 9:16 | 1080x1920 |\\n| Twitter/X Feed | Landscape | 16:9 | 1200x675 |\\n| Twitter/X Card | Landscape | 1.91:1 | 800x418 |\\n\\nInclude these dimensions in your generation prompts to avoid needing to crop or resize.\\n\"},\n {\"skill\":\"ad-creative\",\"ref\":\"platform-specs\",\"body\":\"# Platform Specs Reference\\n\\nComplete character limits, format requirements, and best practices for each ad platform.\\n\\n---\\n\\n## Google Ads\\n\\n### Responsive Search Ads (RSAs)\\n\\n| Element | Character Limit | Required | Notes |\\n|---------|----------------|----------|-------|\\n| Headline | 30 chars | 3 minimum, 15 max | Any 3 may be shown together |\\n| Description | 90 chars | 2 minimum, 4 max | Any 2 may be shown together |\\n| Display path 1 | 15 chars | Optional | Appears after domain in URL |\\n| Display path 2 | 15 chars | Optional | Appears after path 1 |\\n| Final URL | No limit | Required | Landing page URL |\\n\\n**Combination rules:**\\n- Google selects up to 3 headlines and 2 descriptions to show\\n- Headlines appear separated by \\\" | \\\" or stacked\\n- Any headline can appear in any position unless pinned\\n- Pinning reduces Google's ability to optimize — use sparingly\\n\\n**Pinning strategy:**\\n- Pin your brand name to position 1 if brand guidelines require it\\n- Pin your strongest CTA to position 2 or 3\\n- Leave most headlines unpinned for machine learning\\n\\n**Headline mix recommendation (15 headlines):**\\n- 3-4 keyword-focused (match search intent)\\n- 3-4 benefit-focused (what they get)\\n- 2-3 social proof (numbers, awards, customers)\\n- 2-3 CTA-focused (action to take)\\n- 1-2 differentiators (why you over competitors)\\n- 1 brand name headline\\n\\n**Description mix recommendation (4 descriptions):**\\n- 1 benefit + proof point\\n- 1 feature + outcome\\n- 1 social proof + CTA\\n- 1 urgency/offer + CTA (if applicable)\\n\\n### Performance Max\\n\\n| Element | Character Limit | Notes |\\n|---------|----------------|-------|\\n| Headline | 30 chars (5 required) | Short headlines for various placements |\\n| Long headline | 90 chars (5 required) | Used in display, video, discover |\\n| Description | 90 chars (1 required, 5 max) | Accompany various ad formats |\\n| Business name | 25 chars | Required |\\n\\n### Display Ads\\n\\n| Element | Character Limit |\\n|---------|----------------|\\n| Headline | 30 chars |\\n| Long headline | 90 chars |\\n| Description | 90 chars |\\n| Business name | 25 chars |\\n\\n---\\n\\n## Meta Ads (Facebook & Instagram)\\n\\n### Single Image / Video / Carousel\\n\\n| Element | Recommended | Maximum | Notes |\\n|---------|-------------|---------|-------|\\n| Primary text | 125 chars | 2,200 chars | Text above image; truncated after ~125 |\\n| Headline | 40 chars | 255 chars | Below image; truncated after ~40 |\\n| Description | 30 chars | 255 chars | Below headline; may not show |\\n| URL display link | 40 chars | N/A | Optional custom display URL |\\n\\n**Placement-specific notes:**\\n- **Feed**: All elements show; primary text most visible\\n- **Stories/Reels**: Primary text overlaid; keep under 72 chars\\n- **Right column**: Only headline visible; skip description\\n- **Audience Network**: Varies by publisher\\n\\n**Best practices:**\\n- Front-load the hook in primary text (first 125 chars)\\n- Use line breaks for readability in longer primary text\\n- Emojis: test, but don't overuse — 1-2 per ad max\\n- Questions in primary text increase engagement\\n- Headline should be a clear CTA or value statement\\n\\n### Lead Ads (Instant Form)\\n\\n| Element | Limit |\\n|---------|-------|\\n| Greeting headline | 60 chars |\\n| Greeting description | 360 chars |\\n| Privacy policy text | 200 chars |\\n\\n---\\n\\n## LinkedIn Ads\\n\\n### Single Image Ad\\n\\n| Element | Recommended | Maximum | Notes |\\n|---------|-------------|---------|-------|\\n| Intro text | 150 chars | 600 chars | Above the image; truncated after ~150 |\\n| Headline | 70 chars | 200 chars | Below the image |\\n| Description | 100 chars | 300 chars | Only shows on Audience Network |\\n\\n### Carousel Ad\\n\\n| Element | Limit |\\n|---------|-------|\\n| Intro text | 255 chars |\\n| Card headline | 45 chars |\\n| Card count | 2-10 cards |\\n\\n### Message Ad (InMail)\\n\\n| Element | Limit |\\n|---------|-------|\\n| Subject line | 60 chars |\\n| Message body | 1,500 chars |\\n| CTA button | 20 chars |\\n\\n### Text Ad\\n\\n| Element | Limit |\\n|---------|-------|\\n| Headline | 25 chars |\\n| Description | 75 chars |\\n\\n**LinkedIn-specific guidelines:**\\n- Professional tone, but not boring\\n- Use job-specific language the audience recognizes\\n- Statistics and data points perform well\\n- Avoid consumer-style hype (\\\"Amazing!\\\" \\\"Incredible!\\\")\\n- First-person testimonials from peers resonate\\n\\n---\\n\\n## TikTok Ads\\n\\n### In-Feed Ads\\n\\n| Element | Recommended | Maximum | Notes |\\n|---------|-------------|---------|-------|\\n| Ad text | 80 chars | 100 chars | Above the video |\\n| Display name | N/A | 40 chars | Brand name |\\n| CTA button | Platform options | Predefined | Select from TikTok's options |\\n\\n### Spark Ads (Boosted Organic)\\n\\n| Element | Notes |\\n|---------|-------|\\n| Caption | Uses original post caption |\\n| CTA button | Added by advertiser |\\n| Display name | Original creator's handle |\\n\\n**TikTok-specific guidelines:**\\n- Native content outperforms polished ads\\n- First 2 seconds determine if they watch\\n- Use trending sounds and formats\\n- Text overlay is essential (most watch with sound off)\\n- Vertical video only (9:16)\\n\\n---\\n\\n## Twitter/X Ads\\n\\n### Promoted Tweets\\n\\n| Element | Limit | Notes |\\n|---------|-------|-------|\\n| Tweet text | 280 chars | Full tweet with image/video |\\n| Card headline | 70 chars | Website card |\\n| Card description | 200 chars | Website card |\\n\\n### Website Cards\\n\\n| Element | Limit |\\n|---------|-------|\\n| Headline | 70 chars |\\n| Description | 200 chars |\\n\\n**Twitter/X-specific guidelines:**\\n- Conversational, casual tone\\n- Short sentences work best\\n- One clear message per tweet\\n- Hashtags: 1-2 max (0 is often better for ads)\\n- Threads can work for consideration-stage content\\n\\n---\\n\\n## Character Counting Tips\\n\\n- **Spaces count** as characters on all platforms\\n- **Emojis** count as 1-2 characters depending on platform\\n- **Special characters** (|, &, etc.) count as 1 character\\n- **URLs** in body text count against limits\\n- **Dynamic keyword insertion** (`{KeyWord:default}`) can exceed limits — set safe defaults\\n- Always verify in the platform's ad preview before launching\\n\\n---\\n\\n## Multi-Platform Creative Adaptation\\n\\nWhen creating for multiple platforms simultaneously, start with the most restrictive format:\\n\\n1. **Google Search headlines** (30 chars) — forces the tightest messaging\\n2. **Expand to Meta headlines** (40 chars) — add a word or two\\n3. **Expand to LinkedIn intro text** (150 chars) — add context and proof\\n4. **Expand to Meta primary text** (125+ chars) — full hook and value prop\\n\\nThis cascading approach ensures your core message works everywhere, then gets enriched for platforms that allow more space.\\n\"},\n {\"skill\":\"ads\",\"ref\":\"ad-copy-templates\",\"body\":\"# Ad Copy Templates Reference\\n\\nDetailed formulas and templates for writing high-converting ad copy.\\n\\n## Contents\\n- Primary Text Formulas (Problem-Agitate-Solve, Before-After-Bridge, Social Proof Lead, Feature-Benefit Bridge, Direct Response)\\n- Headline Formulas (For Search Ads, For Social Ads)\\n- CTA Variations (Soft CTAs, Hard CTAs, Urgency CTAs, Action-Oriented CTAs)\\n- Platform-Specific Copy Guidelines (Google Search Ads, Meta Ads, LinkedIn Ads)\\n- Copy Testing Priority\\n\\n## Primary Text Formulas\\n\\n### Problem-Agitate-Solve (PAS)\\n\\n```\\n[Problem statement]\\n[Agitate the pain]\\n[Introduce solution]\\n[CTA]\\n```\\n\\n**Example:**\\n> Spending hours on manual reporting every week?\\n> While you're buried in spreadsheets, your competitors are making decisions.\\n> [Product] automates your reports in minutes.\\n> Start your free trial →\\n\\n---\\n\\n### Before-After-Bridge (BAB)\\n\\n```\\n[Current painful state]\\n[Desired future state]\\n[Your product as the bridge]\\n```\\n\\n**Example:**\\n> Before: Chasing down approvals across email, Slack, and spreadsheets.\\n> After: Every approval tracked, automated, and on time.\\n> [Product] connects your tools and keeps projects moving.\\n\\n---\\n\\n### Social Proof Lead\\n\\n```\\n[Impressive stat or testimonial]\\n[What you do]\\n[CTA]\\n```\\n\\n**Example:**\\n> \\\"We cut our reporting time by 75%.\\\" — Sarah K., Marketing Director\\n> [Product] automates the reports you hate building.\\n> See how it works →\\n\\n---\\n\\n### Feature-Benefit Bridge\\n\\n```\\n[Feature]\\n[So that...]\\n[Which means...]\\n```\\n\\n**Example:**\\n> Real-time collaboration on documents\\n> So your team always works from the latest version\\n> Which means no more version confusion or lost work\\n\\n---\\n\\n### Direct Response\\n\\n```\\n[Bold claim/outcome]\\n[Proof point]\\n[CTA with urgency if genuine]\\n```\\n\\n**Example:**\\n> Cut your reporting time by 80%\\n> Join 5,000+ marketing teams already using [Product]\\n> Start free → First month 50% off\\n\\n---\\n\\n## Headline Formulas\\n\\n### For Search Ads\\n\\n| Formula | Example |\\n|---------|---------|\\n| [Keyword] + [Benefit] | \\\"Project Management That Teams Actually Use\\\" |\\n| [Action] + [Outcome] | \\\"Automate Reports \\\\| Save 10 Hours Weekly\\\" |\\n| [Question] | \\\"Tired of Manual Data Entry?\\\" |\\n| [Number] + [Benefit] | \\\"500+ Teams Trust [Product] for [Outcome]\\\" |\\n| [Keyword] + [Differentiator] | \\\"CRM Built for Small Teams\\\" |\\n| [Price/Offer] + [Keyword] | \\\"Free Project Management \\\\| No Credit Card\\\" |\\n\\n### For Social Ads\\n\\n| Type | Example |\\n|------|---------|\\n| Outcome hook | \\\"How we 3x'd our conversion rate\\\" |\\n| Curiosity hook | \\\"The reporting hack no one talks about\\\" |\\n| Contrarian hook | \\\"Why we stopped using [common tool]\\\" |\\n| Specificity hook | \\\"The exact template we use for...\\\" |\\n| Question hook | \\\"What if you could cut your admin time in half?\\\" |\\n| Number hook | \\\"7 ways to improve your workflow today\\\" |\\n| Story hook | \\\"We almost gave up. Then we found...\\\" |\\n\\n---\\n\\n## CTA Variations\\n\\n### Soft CTAs (awareness/consideration)\\n\\nBest for: Top of funnel, cold audiences, complex products\\n\\n- Learn More\\n- See How It Works\\n- Watch Demo\\n- Get the Guide\\n- Explore Features\\n- See Examples\\n- Read the Case Study\\n\\n### Hard CTAs (conversion)\\n\\nBest for: Bottom of funnel, warm audiences, clear offers\\n\\n- Start Free Trial\\n- Get Started Free\\n- Book a Demo\\n- Claim Your Discount\\n- Buy Now\\n- Sign Up Free\\n- Get Instant Access\\n\\n### Urgency CTAs (use when genuine)\\n\\nBest for: Limited-time offers, scarcity situations\\n\\n- Limited Time: 30% Off\\n- Offer Ends [Date]\\n- Only X Spots Left\\n- Last Chance\\n- Early Bird Pricing Ends Soon\\n\\n### Action-Oriented CTAs\\n\\nBest for: Active voice, clear next step\\n\\n- Start Saving Time Today\\n- Get Your Free Report\\n- See Your Score\\n- Calculate Your ROI\\n- Build Your First Project\\n\\n---\\n\\n## Platform-Specific Copy Guidelines\\n\\n### Google Search Ads\\n\\n- **Headline limits:** 30 characters each (up to 15 headlines)\\n- **Description limits:** 90 characters each (up to 4 descriptions)\\n- Include keywords naturally\\n- Use all available headline slots\\n- Include numbers and stats when possible\\n- Test dynamic keyword insertion\\n\\n### Meta Ads (Facebook/Instagram)\\n\\n- **Primary text:** 125 characters visible (can be longer, gets truncated)\\n- **Headline:** 40 characters recommended\\n- Front-load the hook (first line matters most)\\n- Emojis can work but test\\n- Questions perform well\\n- Keep image text under 20%\\n\\n### LinkedIn Ads\\n\\n- **Intro text:** 600 characters max (150 recommended)\\n- **Headline:** 200 characters max (70 recommended)\\n- Professional tone (but not boring)\\n- Specific job outcomes resonate\\n- Stats and social proof important\\n- Avoid consumer-style hype\\n\\n---\\n\\n## Copy Testing Priority\\n\\nWhen testing ad copy, focus on these elements in order of impact:\\n\\n1. **Hook/angle** (biggest impact on performance)\\n2. **Headline**\\n3. **Primary benefit**\\n4. **CTA**\\n5. **Supporting proof points**\\n\\nTest one element at a time for clean data.\\n\"},\n {\"skill\":\"ads\",\"ref\":\"audience-targeting\",\"body\":\"# Audience Targeting Reference\\n\\nDetailed targeting strategies for each major ad platform.\\n\\n## Contents\\n- Google Ads Audiences (Search Campaign Targeting, Display/YouTube Targeting)\\n- Meta Audiences (Core Audiences, Custom Audiences, Lookalike Audiences)\\n- LinkedIn Audiences (Job-Based Targeting, Company-Based Targeting, High-Performing Combinations)\\n- Twitter/X Audiences\\n- TikTok Audiences\\n- Audience Size Guidelines\\n- Exclusion Strategy\\n\\n## Google Ads Audiences\\n\\n### Search Campaign Targeting\\n\\n**Keywords:**\\n- Exact match: [keyword] — most precise, lower volume\\n- Phrase match: \\\"keyword\\\" — moderate precision and volume\\n- Broad match: keyword — highest volume, use with smart bidding\\n\\n**Audience layering:**\\n- Add audiences in \\\"observation\\\" mode first\\n- Analyze performance by audience\\n- Switch to \\\"targeting\\\" mode for high performers\\n\\n**RLSA (Remarketing Lists for Search Ads):**\\n- Bid higher on past visitors searching your terms\\n- Show different ads to returning searchers\\n- Exclude converters from prospecting campaigns\\n\\n### Display/YouTube Targeting\\n\\n**Custom intent audiences:**\\n- Based on recent search behavior\\n- Create from your converting keywords\\n- High intent, good for prospecting\\n\\n**In-market audiences:**\\n- People actively researching solutions\\n- Pre-built by Google\\n- Layer with demographics for precision\\n\\n**Affinity audiences:**\\n- Based on interests and habits\\n- Better for awareness\\n- Broad but can exclude irrelevant\\n\\n**Customer match:**\\n- Upload email lists\\n- Retarget existing customers\\n- Create lookalikes from best customers\\n\\n**Similar/lookalike audiences:**\\n- Based on your customer match lists\\n- Expand reach while maintaining relevance\\n- Best when source list is high-quality customers\\n\\n---\\n\\n## Meta Audiences\\n\\n### Core Audiences (Interest/Demographic)\\n\\n**Interest targeting tips:**\\n- Layer interests with AND logic for precision\\n- Use Audience Insights to research interests\\n- Start broad, let algorithm optimize\\n- Exclude existing customers always\\n\\n**Demographic targeting:**\\n- Age and gender (if product-specific)\\n- Location (down to zip/postal code)\\n- Language\\n- Education and work (limited data now)\\n\\n**Behavior targeting:**\\n- Purchase behavior\\n- Device usage\\n- Travel patterns\\n- Life events\\n\\n### Custom Audiences\\n\\n**Website visitors:**\\n- All visitors (last 180 days max)\\n- Specific page visitors\\n- Time on site thresholds\\n- Frequency (visited X times)\\n\\n**Customer list:**\\n- Upload emails/phone numbers\\n- Match rate typically 30-70%\\n- Refresh regularly for accuracy\\n\\n**Engagement audiences:**\\n- Video viewers (25%, 50%, 75%, 95%)\\n- Page/profile engagers\\n- Form openers\\n- Instagram engagers\\n\\n**App activity:**\\n- App installers\\n- In-app events\\n- Purchase events\\n\\n### Lookalike Audiences\\n\\n**Source audience quality matters:**\\n- Use high-LTV customers, not all customers\\n- Purchasers > leads > all visitors\\n- Minimum 100 source users, ideally 1,000+\\n\\n**Size recommendations:**\\n- 1% — most similar, smallest reach\\n- 1-3% — good balance for most\\n- 3-5% — broader, good for scale\\n- 5-10% — very broad, awareness only\\n\\n**Layering strategies:**\\n- Lookalike + interest = more precision early\\n- Test lookalike-only as you scale\\n- Exclude the source audience\\n\\n---\\n\\n## LinkedIn Audiences\\n\\n### Job-Based Targeting\\n\\n**Job titles:**\\n- Be specific (CMO vs. \\\"Marketing\\\")\\n- LinkedIn normalizes titles, but verify\\n- Stack related titles\\n- Exclude irrelevant titles\\n\\n**Job functions:**\\n- Broader than titles\\n- Combine with seniority level\\n- Good for awareness campaigns\\n\\n**Seniority levels:**\\n- Entry, Senior, Manager, Director, VP, CXO, Partner\\n- Layer with function for precision\\n\\n**Skills:**\\n- Self-reported, less reliable\\n- Good for technical roles\\n- Use as expansion layer\\n\\n### Company-Based Targeting\\n\\n**Company size:**\\n- 1-10, 11-50, 51-200, 201-500, 501-1000, 1001-5000, 5000+\\n- Key filter for B2B\\n\\n**Industry:**\\n- Based on company classification\\n- Can be broad, layer with other criteria\\n\\n**Company names (ABM):**\\n- Upload target account list\\n- Minimum 300 companies recommended\\n- Match rate varies\\n\\n**Company growth rate:**\\n- Hiring rapidly = budget available\\n- Good signal for timing\\n\\n### High-Performing Combinations\\n\\n| Use Case | Targeting Combination |\\n|----------|----------------------|\\n| Enterprise sales | Company size 1000+ + VP/CXO + Industry |\\n| SMB sales | Company size 11-200 + Manager/Director + Function |\\n| Developer tools | Skills + Job function + Company type |\\n| ABM campaigns | Company list + Decision-maker titles |\\n| Broad awareness | Industry + Seniority + Geography |\\n\\n---\\n\\n## Twitter/X Audiences\\n\\n### Targeting options:\\n- Follower lookalikes (accounts similar to followers of X)\\n- Interest categories\\n- Keywords (in tweets)\\n- Conversation topics\\n- Events\\n- Tailored audiences (your lists)\\n\\n### Best practices:\\n- Follower lookalikes of relevant accounts work well\\n- Keyword targeting catches active conversations\\n- Lower CPMs than LinkedIn/Meta\\n- Less precise, better for awareness\\n\\n---\\n\\n## TikTok Audiences\\n\\n### Targeting options:\\n- Demographics (age, gender, location)\\n- Interests (TikTok's categories)\\n- Behaviors (video interactions)\\n- Device (iOS/Android, connection type)\\n- Custom audiences (pixel, customer file)\\n- Lookalike audiences\\n\\n### Best practices:\\n- Younger skew (18-34 primarily)\\n- Interest targeting is broad\\n- Creative matters more than targeting\\n- Let algorithm optimize with broad targeting\\n\\n---\\n\\n## Audience Size Guidelines\\n\\n| Platform | Minimum Recommended | Ideal Range |\\n|----------|-------------------|-------------|\\n| Google Search | 1,000+ searches/mo | 5,000-50,000 |\\n| Google Display | 100,000+ | 500K-5M |\\n| Meta | 100,000+ | 500K-10M |\\n| LinkedIn | 50,000+ | 100K-500K |\\n| Twitter/X | 50,000+ | 100K-1M |\\n| TikTok | 100,000+ | 1M+ |\\n\\nToo narrow = expensive, slow learning\\nToo broad = wasted spend, poor relevance\\n\\n---\\n\\n## Exclusion Strategy\\n\\nAlways exclude:\\n- Existing customers (unless upsell)\\n- Recent converters (7-14 days)\\n- Bounced visitors (<10 sec)\\n- Employees (by company or email list)\\n- Irrelevant page visitors (careers, support)\\n- Competitors (if identifiable)\\n\"}\n];\n","createdAt",["D",1780286224098],"branchVersion","branchId","ecb58646-56c9-11f1-abd8-ee650bb23af1",{"_30":45,"_32":46,"_34":47,"_36":-5,"_38":48,"_40":14,"_41":42,"_24":25},"skills/refs","c9768083-9936-4697-a4d4-80b7978fe701","refs","directory",["D",1780286224098],"id","ecb4ccf6-56c9-11f1-abd8-ee650bb23af1","aihivecommand",["D",1779556068488],"deletedAt","legacyEnvVars","privacy","public","httpPrivacy","description","AIHive Command Center","imageUrl","e8c17657-6210-4c45-bd2b-dbcf2e020293","pinnedFileId","32167c2c-56ca-11f1-955a-ee650bb23af1","pinPreview",true,"pinMigrated","screenshotVersion","likeCount",0,"orgId","019e4de5-f6b1-7488-a017-a3a028bc2ba8","tursoName","ecb4ccf656c911f1abd8ee650bb23af1","tursoHostname","ecb4ccf656c911f1abd8ee650bb23af1-val-town.aws-us-east-2.turso.io","tursoMeasuredAt",["D",1789503703835],"pinned","userPerm","r","valPerm","blobPerm","rw","emailPerm","sqlitePerm","projectPerm","telemetryPerm","user",{"_106":105,"_113":109},"org",{"_106":105,"_107":105,"_108":109,"_18":61,"_110":111,"_112":25},"branch",{"_13":95,"_32":96,"_49":42,"_97":50,"_98":-5,"_99":-5,"_100":-5,"_101":-5,"_102":-5,"_103":51,"_104":105},"isOwned",1111,"main","projectId","forkedBranchId","forkedBranchVersion","forkedProjectPrivacy","forkedProjectId","forkedProjectOrgId","projectName","userHandle","vlduy123","handle","displayName","avatar","https://avatars.githubusercontent.com/u/208025931?v=4","tier","free","canRestrictVals","profileImageUrl","branchName","","adoptOrgId",{"_49":50,"_32":51,"_38":1123,"_53":-5,"_54":25,"_55":56,"_57":56,"_58":59,"_60":-5,"_18":61,"_62":63,"_64":65,"_66":-5,"_67":-5,"_68":69,"_70":71,"_72":73,"_74":75,"_76":1124,"_78":-5,"_79":80,"_81":80,"_82":83,"_84":83,"_85":83,"_86":80,"_87":80,"_88":1125,"_90":1126,"_92":1127,"_94":25},"tree",[386,430,448,461,462,463,464,465,466,467,468,469,470,471,472,473,474,475,476,477,478,479,480,481,482,483,484,485],"files",[386,387,388,389,390,391,392,393,394,395,396,397,398,399,400,401,402,403,404,405,406,407,408,409,410,411,412,413,414,415,416,417,418,419,420,421,422,423,424,425,426,427,428,429,430,431,432,433,434,435,436,437,438,439,440,441,442,443,444,445,446,447,448,449,450,451,452,453,454,455,456,457,458,459,460,461,462,463,464,465,466,467,468,469,470,471,472,473,474,475,476,477,478,479,480,481,482,483,484,485],"counts",{"_375":69,"_376":283,"_377":69,"_378":379,"_380":381},"contents",[136,137,138,139,140,141,142,143,144,145,146,147,148,149,150,151,152,153,154,155,156,157,158,159,160,161,162,163],"upstreamProject","upstreamBranch","emptyEnvVars","flags",[],"hasHttpFiles","grants","orgs","customDomains",{},{"_49":370,"_32":371,"_30":372,"_167":-5,"_41":42,"_40":196,"_169":65,"_38":373,"_13":69,"_34":47,"_36":-5,"_24":25,"_173":-5,"_27":374},{"_49":364,"_32":365,"_30":366,"_167":-5,"_41":42,"_40":367,"_169":65,"_38":368,"_13":69,"_34":47,"_36":-5,"_24":25,"_173":-5,"_27":369},{"_49":358,"_32":359,"_30":360,"_167":-5,"_41":42,"_40":361,"_169":65,"_38":362,"_13":69,"_34":47,"_36":-5,"_24":25,"_173":-5,"_27":363},{"_49":351,"_32":352,"_30":353,"_167":-5,"_41":42,"_40":354,"_169":25,"_38":355,"_13":196,"_34":35,"_36":356,"_24":25,"_173":-5,"_27":357},{"_49":344,"_32":345,"_30":346,"_167":-5,"_41":42,"_40":347,"_169":25,"_38":348,"_13":69,"_34":35,"_36":349,"_24":25,"_173":-5,"_27":350},{"_49":336,"_32":337,"_30":338,"_167":-5,"_41":42,"_40":339,"_169":25,"_38":340,"_13":341,"_34":35,"_36":342,"_24":25,"_173":-5,"_27":343},{"_49":328,"_32":329,"_30":330,"_167":-5,"_41":42,"_40":331,"_169":25,"_38":332,"_13":333,"_34":35,"_36":334,"_24":25,"_173":-5,"_27":335},{"_49":321,"_32":322,"_30":323,"_167":-5,"_41":42,"_40":324,"_169":25,"_38":325,"_13":69,"_34":35,"_36":326,"_24":25,"_173":-5,"_27":327},{"_49":314,"_32":315,"_30":316,"_167":-5,"_41":42,"_40":317,"_169":65,"_38":318,"_13":251,"_34":35,"_36":319,"_24":25,"_173":-5,"_27":320},{"_49":307,"_32":308,"_30":309,"_167":-5,"_41":42,"_40":310,"_169":25,"_38":311,"_13":283,"_34":35,"_36":312,"_24":25,"_173":-5,"_27":313},{"_49":300,"_32":301,"_30":302,"_167":-5,"_41":42,"_40":202,"_169":25,"_38":303,"_13":304,"_34":35,"_36":305,"_24":25,"_173":-5,"_27":306},{"_49":293,"_32":294,"_30":295,"_167":-5,"_41":42,"_40":296,"_169":25,"_38":297,"_13":251,"_34":35,"_36":298,"_24":25,"_173":-5,"_27":299},{"_49":286,"_32":287,"_30":288,"_167":-5,"_41":42,"_40":289,"_169":25,"_38":290,"_13":196,"_34":35,"_36":291,"_24":25,"_173":-5,"_27":292},{"_49":278,"_32":279,"_30":280,"_167":-5,"_41":42,"_40":281,"_169":25,"_38":282,"_13":283,"_34":35,"_36":284,"_24":25,"_173":-5,"_27":285},{"_49":270,"_32":271,"_30":272,"_167":-5,"_41":42,"_40":273,"_169":25,"_38":274,"_13":275,"_34":35,"_36":276,"_24":25,"_173":-5,"_27":277},{"_49":262,"_32":263,"_30":264,"_167":-5,"_41":42,"_40":265,"_169":25,"_38":266,"_13":267,"_34":35,"_36":268,"_24":25,"_173":-5,"_27":269},{"_49":254,"_32":255,"_30":256,"_167":-5,"_41":42,"_40":257,"_169":25,"_38":258,"_13":259,"_34":35,"_36":260,"_24":25,"_173":-5,"_27":261},{"_49":246,"_32":247,"_30":248,"_167":-5,"_41":42,"_40":249,"_169":25,"_38":250,"_13":251,"_34":35,"_36":252,"_24":25,"_173":-5,"_27":253},{"_49":239,"_32":240,"_30":241,"_167":-5,"_41":42,"_40":242,"_169":65,"_38":243,"_13":220,"_34":35,"_36":244,"_24":25,"_173":-5,"_27":245},{"_49":231,"_32":232,"_30":233,"_167":-5,"_41":42,"_40":234,"_169":25,"_38":235,"_13":236,"_34":35,"_36":237,"_24":25,"_173":-5,"_27":238},{"_49":223,"_32":224,"_30":225,"_167":-5,"_41":42,"_40":226,"_169":65,"_38":227,"_13":228,"_34":35,"_36":229,"_24":25,"_173":-5,"_27":230},{"_49":215,"_32":216,"_30":217,"_167":-5,"_41":42,"_40":218,"_169":65,"_38":219,"_13":220,"_34":35,"_36":221,"_24":25,"_173":-5,"_27":222},{"_49":207,"_32":208,"_30":209,"_167":-5,"_41":42,"_40":210,"_169":25,"_38":211,"_13":212,"_34":35,"_36":213,"_24":25,"_173":-5,"_27":214},{"_49":199,"_32":200,"_30":201,"_167":-5,"_41":42,"_40":202,"_169":65,"_38":203,"_13":204,"_34":35,"_36":205,"_24":25,"_173":-5,"_27":206},{"_49":191,"_32":192,"_30":193,"_167":-5,"_41":42,"_40":194,"_169":65,"_38":195,"_13":196,"_34":35,"_36":197,"_24":25,"_173":-5,"_27":198},{"_49":183,"_32":184,"_30":63,"_167":-5,"_41":42,"_40":185,"_169":25,"_38":186,"_13":187,"_34":188,"_36":189,"_24":25,"_173":-5,"_27":190},{"_49":175,"_32":176,"_30":177,"_167":-5,"_41":42,"_40":178,"_169":25,"_38":179,"_13":180,"_34":35,"_36":181,"_24":25,"_173":-5,"_27":182},{"_49":164,"_32":165,"_30":166,"_167":-5,"_41":42,"_40":168,"_169":25,"_38":170,"_13":171,"_34":35,"_36":172,"_24":25,"_173":-5,"_27":174},"72a438d2-5d6a-11f1-b70c-1607ee4eb77e","types.ts","151549be-56ca-11f1-bc41-ee650bb23af1","parentId",639,"latest",["D",1780284719626],15,"export interface Integration {\n id: string;\n name: string; // 'ahrefs'|'ga4'|'gsc'|'linkedin'|'outlook'|'wordpress'|'youtube'|'producthunt'\n api_key?: string;\n api_secret?: string;\n access_token?: string;\n refresh_token?: string;\n extra_config?: string; // JSON string\n connected_at?: string;\n updated_at?: string;\n}\n\nexport interface LLMConfig {\n id: string;\n // QA fix: gemini was already routed in callLLM but missing from the union, causing TS noise.\n provider: 'anthropic' | 'openai' | 'gemini' | 'custom';\n api_key: string;\n endpoint?: string;\n model: string;\n updated_at?: string;\n // Per-provider key presence flags (computed from the 4 api_key_ columns) —\n // used by Settings UI to render \"(saved)\" pills without leaking the actual key.\n has_key?: { anthropic: boolean; openai: boolean; gemini: boolean; custom: boolean };\n}\n\nexport interface Agent {\n id: string;\n name: string;\n workspace: 'marketing' | 'sales';\n description: string;\n system_prompt: string;\n tools: string; // JSON array string\n model?: string;\n enabled: number; // 0 or 1\n created_at: string;\n updated_at: string;\n created_by: 'user' | 'ai';\n // B2: budget cap (USD/month) + running spend, joined from agent_budgets\n monthly_cap?: number;\n current_spend?: number;\n}\n\nexport interface Conversation {\n id: string;\n agent_id?: string;\n workspace: string;\n messages: string; // JSON array string\n created_at: string;\n updated_at: string;\n}\n\nexport interface Task {\n id: string;\n agent_id?: string;\n workspace: string;\n title: string;\n status: 'pending' | 'running' | 'done' | 'failed';\n input?: string;\n output?: string;\n created_at: string;\n completed_at?: string;\n}\n\nexport interface ChatMessage {\n role: 'user' | 'assistant' | 'system' | 'tool';\n content: string;\n timestamp?: string;\n // Tool-use fields (OpenAI-style) carried at runtime by chat.ts / llm.ts.\n tool_calls?: any[];\n tool_call_id?: string;\n reasoning_content?: string;\n}\n","label",[165],"9d2770b4-5d6b-11f1-959f-1607ee4eb77e","router.ts","30479b74-56ca-11f1-a296-ee650bb23af1",641,["D",1780285220417],94,"import { sqlite, initDB, DEFAULT_AGENT_SEEDS } from \"./db.ts\";\nimport { listAgents, createAgent, updateAgent, deleteAgent } from \"./api/agents.ts\";\nimport { getLLMConfig, saveLLMConfig } from \"./api/llm.ts\";\nimport { handleChat, handleChatStream, listConversations, getConversationById, deleteConversation } from \"./api/chat.ts\";\nimport { handleIntegrationProxy, getCredentials, GOOGLE_SERVICES, MICROSOFT_SCOPES } from \"./api/integrations.ts\";\nimport { getAppJS } from \"./frontend_js.ts\";\nimport { getUIJS } from \"./frontend_ui.ts\";\nimport { getChartsJS } from \"./frontend_charts.ts\";\nimport { getRenderersJS } from \"./frontend_renderers.ts\";\nimport { getRenderers2JS } from \"./frontend_renderers2.ts\";\nimport { getCRMJS } from \"./frontend_crm.ts\";\nimport { getPagesJS } from \"./frontend_pages.ts\";\nimport { getCommandJS } from \"./frontend_command.ts\";\nimport { getIntelligenceJS } from \"./frontend_intelligence.ts\";\nimport { getSettingsJS } from \"./frontend_settings.ts\";\nimport { getWidgetsJS } from \"./frontend_widgets.ts\";\nimport { getObjectivesJS } from \"./frontend_objectives.ts\";\nimport { getCalendarJS } from \"./frontend_calendar.ts\";\nimport { getExperimentsJS } from \"./frontend_experiments.ts\";\nimport { getFunnelSummary, getCorrelations, getRecommendations, getLeadScores, getAIOKpis, getIntelligenceFeed } from \"./api/intelligence.ts\";\nimport { listSignals, recordSignal, actionSignal, getHotLeads, getHotLeadsCount } from \"./api/signals.ts\";\nimport { listObjectives, createObjective, updateObjective, deleteObjective, createKeyResult, updateKeyResult, getOKRDashboard, syncObjectivesFromFunnel } from \"./api/okrs.ts\";\nimport { getNSM, setNSM, recordNSMValue, getNSMProgress } from \"./api/nsm.ts\";\nimport { listCohorts, createCohort, deleteCohort, getCohortSummary, getEngagementVelocity, getAttributionSummary, recordConversion } from \"./api/aio.ts\";\nimport { getLatestDigest, generateDigest, getDigestHistory } from \"./api/digest.ts\";\nimport { listExperiments, createExperiment, updateExperiment, deleteExperiment, runExperiment, completeExperiment, getExperimentStats, suggestICE } from \"./api/experiments.ts\";\nimport { getUnifiedCalendar, getCalendarSummary, createCalendarEntry, updateCalendarEntry, deleteCalendarEntry } from \"./api/calendar_unified.ts\";\nimport { getGraphify } from \"./api/graphify.ts\";\nimport { listTasks, createTask, updateTask, deleteTask } from \"./api/tasks.ts\";\nimport { listDeals, getDeal, createDeal, updateDeal, deleteDeal, syncDealsFromLeads, listCandidates, importDeals } from \"./api/deals.ts\";\nimport { listKnowledgeFiles, getKnowledgeFile, createKnowledgeFile, updateKnowledgeFile, deleteKnowledgeFile, summarizeKnowledgeFile, buildKnowledgeGraph, AI_FEATURES } from \"./api/knowledge.ts\";\nimport { getCompanyProfile, setCompanyProfile, getBrandContext, setBrandContext } from \"./api/company.ts\";\nimport { listMemory, addMemory, updateMemory, deleteMemory } from \"./api/memory.ts\";\nimport { getCampaignsJS } from \"./frontend_campaigns.ts\";\nimport { getKnowledgeGraphJS } from \"./frontend_knowledge_graph.ts\";\nimport { getApprovalsJS } from \"./frontend_approvals.ts\";\nimport { getMethodConfig, saveMethodConfig, getMethodsMetadata } from \"./api/method_config.ts\";\nimport { listActions, enqueueAction, getAction, approveAction, rejectAction } from \"./api/actions.ts\";\nimport { processPendingRuns } from \"./cron.ts\";\nimport { getExecutedWorkCounts } from \"./api/propagate.ts\";\n\nconst corsHeaders = {\n \"Access-Control-Allow-Origin\": \"*\",\n \"Access-Control-Allow-Methods\": \"GET, POST, PUT, DELETE, OPTIONS\",\n \"Access-Control-Allow-Headers\": \"Content-Type, Authorization\",\n};\n\nfunction json(data: unknown, status = 200): Response {\n return new Response(JSON.stringify(data), {\n status,\n headers: {\n \"Content-Type\": \"application/json\",\n ...corsHeaders,\n },\n });\n}\n\n// Mask an integration row before returning it to the client. NO endpoint (list OR upsert)\n// may return secrets in cleartext — only presence flags + a short masked hint.\nfunction maskIntegration(obj: any, googleConnected = false) {\n return {\n ...obj,\n connected: !!(obj.api_key || obj.access_token) ||\n (GOOGLE_SERVICES.includes(String(obj.name)) && googleConnected),\n has_api_key: !!obj.api_key,\n has_api_secret: !!obj.api_secret,\n has_access_token: !!obj.access_token,\n api_key: obj.api_key ? String(obj.api_key).slice(0, 4) + \"***\" : undefined,\n api_secret: obj.api_secret ? \"***\" : undefined,\n access_token: obj.access_token ? \"***\" : undefined,\n refresh_token: obj.refresh_token ? \"***\" : undefined,\n };\n}\n\n// --- OAuth CSRF state (cookie-bound) -----------------------------------------\n// On the auth-start request we mint a random state, echo it in the provider URL, and\n// store it in an HttpOnly cookie. The callback must present a matching state or we reject —\n// this prevents login-CSRF / token fixation into the shared single-tenant credential row.\nfunction newOAuthState(): { state: string; setCookie: string } {\n const state = crypto.randomUUID();\n return { state, setCookie: `oauth_state=${state}; Path=/; HttpOnly; SameSite=Lax; Max-Age=600` };\n}\nconst CLEAR_OAUTH_STATE = \"oauth_state=; Path=/; HttpOnly; SameSite=Lax; Max-Age=0\";\nfunction readCookie(req: Request, name: string): string {\n const raw = req.headers.get(\"cookie\") || \"\";\n const m = raw.match(new RegExp(\"(?:^|;\\\\s*)\" + name + \"=([^;]+)\"));\n return m ? m[1] : \"\";\n}\nfunction redirectWithCookie(location: string, setCookie?: string): Response {\n const headers: Record = { Location: location };\n if (setCookie) headers[\"Set-Cookie\"] = setCookie;\n return new Response(null, { status: 302, headers });\n}\nfunction oauthStateValid(req: Request, url: URL): boolean {\n const expected = readCookie(req, \"oauth_state\");\n const got = url.searchParams.get(\"state\") || \"\";\n return !!expected && !!got && expected === got;\n}\n\nexport async function handleRequest(req: Request, url: URL): Promise {\n const { pathname } = url;\n const method = req.method;\n\n // CORS preflight\n if (method === \"OPTIONS\") {\n return new Response(null, { status: 204, headers: corsHeaders });\n }\n\n try {\n // GET /api/health\n if (pathname === \"/api/health\" && method === \"GET\") {\n return json({ status: \"ok\", timestamp: new Date().toISOString() });\n }\n\n // GET /api/health/auth - gated token verifier. Reaching this (past the main.ts gate)\n // means the supplied APP_TOKEN is valid; the login screen uses it to confirm the password.\n if (pathname === \"/api/health/auth\" && method === \"GET\") {\n return json({ ok: true });\n }\n\n // GET /api/config - LLM runtime config + ALL app_config key/values (workspace_prefs, ROI inputs, …)\n if (pathname === \"/api/config\" && method === \"GET\") {\n const llmConfig = await getLLMConfig();\n const merged: Record = {\n provider: llmConfig.provider,\n model: llmConfig.model,\n has_key: (llmConfig as any).has_key,\n };\n try {\n const r = await sqlite.execute(\"SELECT key, value FROM app_config\");\n for (const row of (r.rows || []) as any[]) {\n const k = String(row[0]);\n let v: any = row[1];\n if (typeof v === \"string\") { try { v = JSON.parse(v); } catch { /* keep raw string */ } }\n merged[k] = v;\n }\n } catch { /* app_config missing → just return LLM fields */ }\n return json(merged);\n }\n // POST /api/config - upsert arbitrary key/value into app_config (workspace prefs, ROI inputs, temperature…).\n // Without this, Settings toggles/ROI inputs silently failed to persist (POST was 404).\n if (pathname === \"/api/config\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n const updated: Record = {};\n try {\n for (const [k, v] of Object.entries(body)) {\n if (k === \"provider\" || k === \"model\" || k === \"has_key\") continue;\n const value = typeof v === \"string\" ? v : JSON.stringify(v);\n await sqlite.execute(\n \"INSERT INTO app_config (key, value) VALUES (?, ?) ON CONFLICT(key) DO UPDATE SET value = excluded.value, updated_at = datetime('now')\",\n [k, value],\n );\n updated[k] = v;\n }\n return json({ success: true, updated });\n } catch (e: any) {\n return json({ error: String(e?.message || e) }, 500);\n }\n }\n\n // --- Frontend JS module routes --------------------------------------------\n const jsModules: Record string> = {\n \"/api/app-js\": getAppJS,\n \"/api/app-ui\": getUIJS,\n \"/api/app-charts\": getChartsJS,\n \"/api/app-renderers\": getRenderersJS,\n \"/api/app-renderers2\": getRenderers2JS,\n \"/api/app-pages\": getPagesJS,\n \"/api/app-command\": getCommandJS,\n \"/api/app-intelligence\": getIntelligenceJS,\n \"/api/app-settings\": getSettingsJS,\n \"/api/app-widgets\": getWidgetsJS,\n \"/api/app-objectives\": getObjectivesJS,\n \"/api/app-calendar\": getCalendarJS,\n \"/api/app-experiments\": getExperimentsJS,\n \"/api/app-campaigns\": getCampaignsJS,\n \"/api/app-knowledge-graph\": getKnowledgeGraphJS,\n \"/api/app-approvals\": getApprovalsJS,\n \"/api/app-crm\": getCRMJS,\n };\n if (jsModules[pathname] && method === \"GET\") {\n return new Response(jsModules[pathname](), {\n headers: {\n \"Content-Type\": \"application/javascript\",\n \"Cache-Control\": \"no-cache, no-store, must-revalidate\",\n ...corsHeaders,\n },\n });\n }\n\n // --- Intelligence routes --------------------------------------------------\n if (pathname === \"/api/intelligence/funnel-summary\" && method === \"GET\") {\n const force = url.searchParams.get(\"force\") === \"1\";\n const days = parseInt(url.searchParams.get(\"days\") ?? \"30\", 10);\n return json(await getFunnelSummary(force, days));\n }\n if (pathname === \"/api/intelligence/correlations\" && method === \"GET\") {\n const force = url.searchParams.get(\"force\") === \"1\";\n return json(await getCorrelations(force));\n }\n if (pathname === \"/api/intelligence/recommendations\" && method === \"GET\") {\n return json(await getRecommendations());\n }\n if (pathname === \"/api/intelligence/leads\" && method === \"GET\") {\n return json(await getLeadScores());\n }\n if (pathname === \"/api/intelligence/kpis\" && method === \"GET\") {\n return json(await getAIOKpis());\n }\n if (pathname === \"/api/intelligence/feed\" && method === \"GET\") {\n return json(await getIntelligenceFeed());\n }\n // Execution-aware metric: how much real work was completed (tasks/experiments) — so Measure\n // reflects execution, not just external integration numbers.\n if (pathname === \"/api/intelligence/executed-work\" && method === \"GET\") {\n const days = parseInt(url.searchParams.get(\"days\") ?? \"30\", 10);\n return json(await getExecutedWorkCounts(days));\n }\n\n // --- Signals routes -------------------------------------------------------\n if (pathname === \"/api/signals\" && method === \"GET\") {\n const actioned = url.searchParams.get(\"actioned\");\n const lead_email = url.searchParams.get(\"lead_email\") ?? undefined;\n return json(await listSignals({ lead_email, actioned: actioned === \"true\" ? true : actioned === \"false\" ? false : undefined }));\n }\n if (pathname === \"/api/signals\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n if (!body.lead_email || !body.signal_type) {\n return json({ error: \"lead_email and signal_type are required\" }, 400);\n }\n return json(await recordSignal(body), 201);\n }\n const signalActionMatch = pathname.match(/^\\/api\\/signals\\/([^/]+)\\/action$/);\n if (signalActionMatch && method === \"PUT\") {\n const result = await actionSignal(signalActionMatch[1]);\n return result ? json(result) : json({ error: \"Signal not found\" }, 404);\n }\n if (pathname === \"/api/signals/hot-leads\" && method === \"GET\") {\n return json(await getHotLeads());\n }\n if (pathname === \"/api/signals/hot-leads/count\" && method === \"GET\") {\n return json({ count: await getHotLeadsCount() });\n }\n\n // --- OKR / Objectives routes ----------------------------------------------\n if (pathname === \"/api/okrs\" && method === \"GET\") {\n const workspace = url.searchParams.get(\"workspace\") ?? undefined;\n return json(await listObjectives({ workspace }));\n }\n if (pathname === \"/api/okrs\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await createObjective(body), 201);\n }\n if (pathname === \"/api/okrs/dashboard\" && method === \"GET\") {\n return json(await getOKRDashboard());\n }\n // Explicit funnel→OKR sync (persists KR/objective progress). The GET list no longer\n // writes on read, so this is the one place that refreshes stored progress on demand.\n if (pathname === \"/api/okrs/sync\" && method === \"POST\") {\n const workspace = url.searchParams.get(\"workspace\") ?? undefined;\n return json(await syncObjectivesFromFunnel({ workspace }));\n }\n const okrIdMatch = pathname.match(/^\\/api\\/okrs\\/([^/]+)$/);\n if (okrIdMatch && method === \"PUT\") {\n const body = await req.json().catch(() => ({}));\n return json(await updateObjective(okrIdMatch[1], body));\n }\n if (okrIdMatch && method === \"DELETE\") {\n const ok = await deleteObjective(okrIdMatch[1]);\n return ok ? json({ success: true }) : json({ error: \"Not found\" }, 404);\n }\n const okrKrMatch = pathname.match(/^\\/api\\/okrs\\/([^/]+)\\/key-results$/);\n if (okrKrMatch && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await createKeyResult(okrKrMatch[1], body), 201);\n }\n const krIdMatch = pathname.match(/^\\/api\\/okrs\\/key-results\\/([^/]+)$/);\n if (krIdMatch && method === \"PUT\") {\n const body = await req.json().catch(() => ({}));\n return json(await updateKeyResult(krIdMatch[1], body));\n }\n\n // --- NSM routes -----------------------------------------------------------\n if (pathname === \"/api/nsm\" && method === \"GET\") {\n return json(await getNSM());\n }\n if (pathname === \"/api/nsm\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await setNSM(body));\n }\n if (pathname === \"/api/nsm/record\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n await recordNSMValue(body.value);\n return json({ success: true });\n }\n if (pathname === \"/api/nsm/progress\" && method === \"GET\") {\n return json(await getNSMProgress());\n }\n\n // --- AIO / Cohort routes --------------------------------------------------\n if (pathname === \"/api/aio/cohorts\" && method === \"GET\") {\n return json(await listCohorts());\n }\n if (pathname === \"/api/aio/cohorts\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await createCohort(body), 201);\n }\n const cohortIdMatch = pathname.match(/^\\/api\\/aio\\/cohorts\\/([^/]+)$/);\n if (cohortIdMatch && method === \"DELETE\") {\n return json({ success: await deleteCohort(cohortIdMatch[1]) });\n }\n if (pathname === \"/api/aio/cohort-summary\" && method === \"GET\") {\n return json(await getCohortSummary());\n }\n if (pathname === \"/api/aio/engagement-velocity\" && method === \"GET\") {\n return json(await getEngagementVelocity());\n }\n\n // --- Frontend-expected aliases (cohorts/attribution/key-results) ----------\n if (pathname === \"/api/cohorts\" && method === \"GET\") {\n return json(await getCohortSummary());\n }\n if (pathname === \"/api/cohorts\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await createCohort(body), 201);\n }\n const cohortAliasMatch = pathname.match(/^\\/api\\/cohorts\\/([^/]+)$/);\n if (cohortAliasMatch && method === \"DELETE\") {\n return json({ success: await deleteCohort(cohortAliasMatch[1]) });\n }\n if (pathname === \"/api/attribution\" && method === \"GET\") {\n return json(await getAttributionSummary());\n }\n if (pathname === \"/api/attribution\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n await recordConversion(body);\n return json({ success: true }, 201);\n }\n const krAliasMatch = pathname.match(/^\\/api\\/key-results\\/([^/]+)$/);\n if (krAliasMatch && method === \"PUT\") {\n const body = await req.json().catch(() => ({}));\n return json(await updateKeyResult(krAliasMatch[1], body));\n }\n\n // --- Digest routes --------------------------------------------------------\n if (pathname === \"/api/digest\" && method === \"GET\") {\n return json(await getLatestDigest());\n }\n if (pathname === \"/api/digest/generate\" && method === \"POST\") {\n return json(await generateDigest());\n }\n if (pathname === \"/api/digest/history\" && method === \"GET\") {\n const limit = parseInt(url.searchParams.get(\"limit\") ?? \"10\", 10);\n return json(await getDigestHistory(limit));\n }\n\n // --- Experiments routes ---------------------------------------------------\n if (pathname === \"/api/experiments\" && method === \"GET\") {\n const status = url.searchParams.get(\"status\") ?? undefined;\n const funnel_stage = url.searchParams.get(\"stage\") ?? url.searchParams.get(\"funnel_stage\") ?? undefined;\n const aarrr_lever = url.searchParams.get(\"aarrr_lever\") ?? undefined;\n return json(await listExperiments({ status, funnel_stage, aarrr_lever }));\n }\n if (pathname === \"/api/experiments\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await createExperiment(body), 201);\n }\n if (pathname === \"/api/experiments/stats\" && method === \"GET\") {\n return json(await getExperimentStats());\n }\n if (pathname === \"/api/experiments/ice-suggest\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await suggestICE(body));\n }\n const expActionMatch = pathname.match(/^\\/api\\/experiments\\/([^/]+)\\/(run|complete)$/);\n if (expActionMatch && method === \"POST\") {\n const id = expActionMatch[1];\n const body = await req.json().catch(() => ({}));\n const result = expActionMatch[2] === \"run\"\n ? await runExperiment(id, Number(body.baseline_value) || 0)\n : await completeExperiment(id, {\n result_value: Number(body.result_value) || 0,\n outcome: body.outcome ?? \"inconclusive\",\n learnings: body.learnings,\n });\n return result ? json(result) : json({ error: \"Experiment not found\" }, 404);\n }\n const expIdMatch = pathname.match(/^\\/api\\/experiments\\/([^/]+)$/);\n if (expIdMatch && method === \"PUT\") {\n const body = await req.json().catch(() => ({}));\n const updated = await updateExperiment(expIdMatch[1], body);\n return updated ? json(updated) : json({ error: \"Experiment not found\" }, 404);\n }\n if (expIdMatch && method === \"DELETE\") {\n const ok = await deleteExperiment(expIdMatch[1]);\n return ok ? json({ success: true }) : json({ error: \"Not found\" }, 404);\n }\n\n // --- Calendar unified route -----------------------------------------------\n if (pathname === \"/api/calendar/unified\" && method === \"GET\") {\n const today = new Date().toISOString().slice(0, 10);\n const from = url.searchParams.get(\"from\") ?? url.searchParams.get(\"start\") ?? today;\n const to = url.searchParams.get(\"to\") ?? url.searchParams.get(\"end\") ?? from;\n const workspace = url.searchParams.get(\"workspace\") ?? undefined;\n const sourcesParam = url.searchParams.getAll(\"source\");\n return json(await getUnifiedCalendar({ from, to, workspace, sources: sourcesParam.length ? sourcesParam as any : undefined }));\n }\n if (pathname === \"/api/calendar/summary\" && method === \"GET\") {\n const date = url.searchParams.get(\"date\") ?? new Date().toISOString().slice(0, 10);\n return json(await getCalendarSummary(date));\n }\n if (pathname === \"/api/calendar/entries\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await createCalendarEntry(body), 201);\n }\n const calEntryMatch = pathname.match(/^\\/api\\/calendar\\/entries\\/([^/]+)$/);\n if (calEntryMatch && method === \"PUT\") {\n const body = await req.json().catch(() => ({}));\n const updated = await updateCalendarEntry(calEntryMatch[1], body);\n return updated ? json(updated) : json({ error: \"Not found\" }, 404);\n }\n if (calEntryMatch && method === \"DELETE\") {\n const ok = await deleteCalendarEntry(calEntryMatch[1]);\n return ok ? json({ success: true }) : json({ error: \"Not found\" }, 404);\n }\n\n // --- Knowledge files (AI knowledge injection + graph view) ----------------\n if (pathname === \"/api/knowledge\" && method === \"GET\") {\n return json(await listKnowledgeFiles());\n }\n if (pathname === \"/api/knowledge\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n const created = await createKnowledgeFile(body);\n return created ? json(created, 201) : json({ error: \"content required\" }, 400);\n }\n if (pathname === \"/api/knowledge/graph\" && method === \"GET\") {\n return json(await buildKnowledgeGraph());\n }\n if (pathname === \"/api/knowledge/features\" && method === \"GET\") {\n return json(AI_FEATURES);\n }\n const knSummarizeMatch = pathname.match(/^\\/api\\/knowledge\\/([^/]+)\\/summarize$/);\n if (knSummarizeMatch && method === \"POST\") {\n const summary = await summarizeKnowledgeFile(knSummarizeMatch[1]);\n const file = await getKnowledgeFile(knSummarizeMatch[1]);\n return json({ summary, tokens_estimate: file?.tokens_estimate ?? 0 });\n }\n const knIdMatch = pathname.match(/^\\/api\\/knowledge\\/([^/]+)$/);\n if (knIdMatch && method === \"GET\") {\n const f = await getKnowledgeFile(knIdMatch[1]);\n return f ? json(f) : json({ error: \"Not found\" }, 404);\n }\n if (knIdMatch && method === \"PUT\") {\n const body = await req.json().catch(() => ({}));\n const updated = await updateKnowledgeFile(knIdMatch[1], body);\n return updated ? json(updated) : json({ error: \"Not found\" }, 404);\n }\n if (knIdMatch && method === \"DELETE\") {\n const ok = await deleteKnowledgeFile(knIdMatch[1]);\n return ok ? json({ success: true }) : json({ error: \"Not found\" }, 404);\n }\n\n // --- Company Profile + Brand Context (Part 15.2) — the business identity ---\n if (pathname === \"/api/company-profile\" && method === \"GET\") {\n return json(await getCompanyProfile());\n }\n if (pathname === \"/api/company-profile\" && method === \"PUT\") {\n const body = await req.json().catch(() => ({}));\n return json(await setCompanyProfile(body));\n }\n if (pathname === \"/api/brand-context\" && method === \"GET\") {\n return json({ text: await getBrandContext() });\n }\n if (pathname === \"/api/brand-context\" && method === \"PUT\") {\n const body = await req.json().catch(() => ({}));\n await setBrandContext(String(body.text ?? \"\"));\n return json({ ok: true });\n }\n\n // --- Memory (Part 15.3) — persistent evolving business facts --------------\n if (pathname === \"/api/memory\" && method === \"GET\") {\n return json(await listMemory());\n }\n if (pathname === \"/api/memory\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n const m = await addMemory(body);\n return m ? json(m, 201) : json({ error: \"content required\" }, 400);\n }\n const memIdMatch = pathname.match(/^\\/api\\/memory\\/([0-9]+)$/);\n if (memIdMatch && method === \"PUT\") {\n const body = await req.json().catch(() => ({}));\n const m = await updateMemory(Number(memIdMatch[1]), body);\n return m ? json(m) : json({ error: \"Not found\" }, 404);\n }\n if (memIdMatch && method === \"DELETE\") {\n const ok = await deleteMemory(Number(memIdMatch[1]));\n return ok ? json({ success: true }) : json({ error: \"Not found\" }, 404);\n }\n\n // --- Method config (prioritization methodology: ICE/RICE/WSJF) ------------\n // Backend module existed but was unrouted → the Settings methodology panel 404'd.\n if (pathname === \"/api/method-config/methods\" && method === \"GET\") {\n return json(getMethodsMetadata());\n }\n if (pathname === \"/api/method-config\" && method === \"GET\") {\n return json(await getMethodConfig());\n }\n if (pathname === \"/api/method-config\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await saveMethodConfig(body));\n }\n\n // --- Action queue (human-in-the-loop execution) ---------------------------\n if (pathname === \"/api/actions\" && method === \"GET\") {\n const status = url.searchParams.get(\"status\") ?? undefined;\n return json(await listActions(status));\n }\n if (pathname === \"/api/actions\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n const created = await enqueueAction(body);\n return created ? json(created, 201) : json({ error: \"integration + endpoint required\" }, 400);\n }\n const actionApproveMatch = pathname.match(/^\\/api\\/actions\\/([^/]+)\\/approve$/);\n if (actionApproveMatch && method === \"POST\") {\n const a = await approveAction(actionApproveMatch[1]);\n return a ? json(a) : json({ error: \"Not found\" }, 404);\n }\n const actionRejectMatch = pathname.match(/^\\/api\\/actions\\/([^/]+)\\/reject$/);\n if (actionRejectMatch && method === \"POST\") {\n const a = await rejectAction(actionRejectMatch[1]);\n return a ? json(a) : json({ error: \"Not found\" }, 404);\n }\n const actionIdMatch = pathname.match(/^\\/api\\/actions\\/([^/]+)$/);\n if (actionIdMatch && method === \"GET\") {\n const a = await getAction(actionIdMatch[1]);\n return a ? json(a) : json({ error: \"Not found\" }, 404);\n }\n\n // Manual trigger for the execution engine (the interval val also calls processPendingRuns).\n if (pathname === \"/api/agents/run-pending\" && method === \"POST\") {\n const limit = parseInt(url.searchParams.get(\"limit\") ?? \"3\", 10);\n return json(await processPendingRuns(limit));\n }\n\n // --- Agent runs -----------------------------------------------------------\n const agentRunsMatch = pathname.match(/^\\/api\\/agents\\/([^/]+)\\/runs$/);\n if (agentRunsMatch && method === \"GET\") {\n const agent_id = agentRunsMatch[1];\n const limit = parseInt(url.searchParams.get(\"limit\") ?? \"20\", 10);\n const result = await sqlite.execute(\n \"SELECT * FROM agent_runs WHERE agent_id = ? ORDER BY started_at DESC LIMIT ?\",\n [agent_id, limit]\n );\n return json(result.rows.map((row) => rowToObject(result.columns, row)));\n }\n if (agentRunsMatch && method === \"POST\") {\n const agent_id = agentRunsMatch[1];\n const body = await req.json().catch(() => ({}));\n const id = crypto.randomUUID();\n const now = new Date().toISOString();\n await sqlite.execute(\n \"INSERT INTO agent_runs (id, agent_id, trigger, status, started_at) VALUES (?, ?, ?, ?, ?)\",\n [id, agent_id, body.trigger ?? \"manual\", \"pending\", now]\n );\n return json({ id, agent_id, trigger: body.trigger ?? \"manual\", status: \"pending\", started_at: now }, 201);\n }\n\n // Integrations routes\n // GET /api/integrations - list all integrations\n if (pathname === \"/api/integrations\" && method === \"GET\") {\n const result = await sqlite.execute(\"SELECT * FROM integrations ORDER BY name ASC\");\n const rows = result.rows.map((row) => rowToObject(result.columns, row));\n // Google-family services inherit the central 'google' OAuth token, so they are \"connected\"\n // whenever google is — even without their own api_key/access_token row.\n const googleConnected = rows.some((o) => o.name === \"google\" && o.access_token);\n const integrations = rows.map((obj) => maskIntegration(obj, googleConnected));\n return json(integrations);\n }\n\n // POST /api/integrations/:name - upsert integration credentials\n const integrationUpsertMatch = pathname.match(/^\\/api\\/integrations\\/([^/]+)$/);\n if (integrationUpsertMatch && method === \"POST\") {\n const name = integrationUpsertMatch[1];\n const body = await req.json().catch(() => ({}));\n const now = new Date().toISOString();\n\n const existing = await sqlite.execute(\n \"SELECT id, api_key, api_secret, access_token, refresh_token, extra_config FROM integrations WHERE name = ?\",\n [name]\n );\n\n // Preserve an existing secret when the incoming value is blank, missing, or a masked\n // placeholder (\"***\"). Saving the (masked) Settings form must NOT wipe stored credentials —\n // this is what silently cleared the Ahrefs key.\n const keepSecret = (incoming: any, current: any) => {\n if (incoming == null) return current ?? null;\n const s = String(incoming);\n if (s === \"\" || s.includes(\"***\")) return current ?? null;\n return incoming;\n };\n\n if (existing.rows.length > 0) {\n const cur = existing.rows[0] as any[]; // [id, api_key, api_secret, access_token, refresh_token, extra_config]\n await sqlite.execute(\n `UPDATE integrations SET\n api_key = ?,\n api_secret = ?,\n access_token = ?,\n refresh_token = ?,\n extra_config = ?,\n updated_at = ?\n WHERE name = ?`,\n [\n keepSecret(body.api_key, cur[1]),\n keepSecret(body.api_secret, cur[2]),\n keepSecret(body.access_token, cur[3]),\n keepSecret(body.refresh_token, cur[4]),\n body.extra_config != null ? JSON.stringify(body.extra_config) : (cur[5] ?? null),\n now,\n name,\n ]\n );\n } else {\n // Insert\n const id = crypto.randomUUID();\n await sqlite.execute(\n `INSERT INTO integrations (id, name, api_key, api_secret, access_token, refresh_token, extra_config, connected_at, updated_at)\n VALUES (?, ?, ?, ?, ?, ?, ?, ?, ?)`,\n [\n id,\n name,\n body.api_key ?? null,\n body.api_secret ?? null,\n body.access_token ?? null,\n body.refresh_token ?? null,\n body.extra_config != null ? JSON.stringify(body.extra_config) : null,\n now,\n now,\n ]\n );\n }\n\n const updated = await sqlite.execute(\n \"SELECT * FROM integrations WHERE name = ?\",\n [name]\n );\n const raw = rowToObject(updated.columns, updated.rows[0]);\n // NEVER return the just-saved secrets in cleartext — mask like the list endpoint.\n const googleConnected = raw.name === \"google\"\n ? !!raw.access_token\n : !!(await getCredentials(\"google\"))?.access_token;\n return json({ success: true, integration: maskIntegration(raw, googleConnected) });\n }\n\n // DELETE /api/integrations/:name\n const integrationDeleteMatch = pathname.match(/^\\/api\\/integrations\\/([^/]+)$/);\n if (integrationDeleteMatch && method === \"DELETE\") {\n const name = integrationDeleteMatch[1];\n await sqlite.execute(\"DELETE FROM integrations WHERE name = ?\", [name]);\n return json({ success: true, message: `Integration '${name}' deleted` });\n }\n\n // GET|POST /api/integrations/:name/data - proxy to integration API\n const integrationDataMatch = pathname.match(/^\\/api\\/integrations\\/([^/]+)\\/data$/);\n if (integrationDataMatch && (method === \"GET\" || method === \"POST\")) {\n const name = integrationDataMatch[1];\n const credentials = await getCredentials(name);\n if (!credentials) {\n return json({ error: `Integration '${name}' not found or not configured` }, 404);\n }\n return handleIntegrationProxy(req, name, url.searchParams, credentials);\n }\n\n // Agents routes\n // GET /api/agents - list agents (optional ?workspace= query param)\n if (pathname === \"/api/agents\" && method === \"GET\") {\n const workspace = url.searchParams.get(\"workspace\") ?? undefined;\n const agents = await listAgents(workspace);\n return json(agents);\n }\n\n // POST /api/agents - create agent\n if (pathname === \"/api/agents\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n const agent = await createAgent(body);\n return json(agent, 201);\n }\n\n // PUT /api/agents/:id - update agent\n const agentUpdateMatch = pathname.match(/^\\/api\\/agents\\/([^/]+)$/);\n if (agentUpdateMatch && method === \"PUT\") {\n const id = agentUpdateMatch[1];\n const body = await req.json().catch(() => ({}));\n const agent = await updateAgent(id, body);\n if (!agent) {\n return json({ error: `Agent '${id}' not found` }, 404);\n }\n return json(agent);\n }\n\n // DELETE /api/agents/:id - delete agent\n const agentDeleteMatch = pathname.match(/^\\/api\\/agents\\/([^/]+)$/);\n if (agentDeleteMatch && method === \"DELETE\") {\n const id = agentDeleteMatch[1];\n const deleted = await deleteAgent(id);\n if (!deleted) {\n return json({ error: `Agent '${id}' not found` }, 404);\n }\n return json({ success: true, message: `Agent '${id}' deleted` });\n }\n\n // LLM config routes\n // GET /api/llm/config - get LLM config (mask api_key)\n if (pathname === \"/api/llm/config\" && method === \"GET\") {\n const config = await getLLMConfig();\n if (!config) {\n return json({ error: \"LLM config not found\" }, 404);\n }\n // Mask api_key: show first 8 chars + ***\n const masked = {\n ...config,\n api_key: config.api_key\n ? config.api_key.slice(0, 8) + \"***\"\n : \"\",\n };\n return json(masked);\n }\n\n // POST /api/llm/config - save LLM config\n if (pathname === \"/api/llm/config\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n const config = await saveLLMConfig(body);\n // Mask api_key in response\n const masked = {\n ...config,\n api_key: config.api_key\n ? config.api_key.slice(0, 8) + \"***\"\n : \"\",\n };\n return json(masked);\n }\n\n // Chat route\n // POST /api/chat - AI chat handler\n if (pathname === \"/api/chat\" && method === \"POST\") {\n return handleChat(req);\n }\n\n // POST /api/chat/stream - SSE streaming chat\n if (pathname === \"/api/chat/stream\" && method === \"POST\") {\n return handleChatStream(req);\n }\n\n // --- Conversation history routes ------------------------------------------\n if (pathname === \"/api/conversations\" && method === \"GET\") {\n return json(await listConversations());\n }\n const convIdMatch = pathname.match(/^\\/api\\/conversations\\/([^/]+)$/);\n if (convIdMatch && method === \"GET\") {\n const conv = await getConversationById(convIdMatch[1]);\n return conv ? json(conv) : json({ error: \"Not found\" }, 404);\n }\n if (convIdMatch && method === \"DELETE\") {\n const ok = await deleteConversation(convIdMatch[1]);\n return ok ? json({ success: true }) : json({ error: \"Not found\" }, 404);\n }\n\n // Tasks routes\n // GET /api/tasks - list tasks (optional ?workspace= query param)\n if (pathname === \"/api/tasks\" && method === \"GET\") {\n const workspace = url.searchParams.get(\"workspace\") ?? undefined;\n const status = url.searchParams.get(\"status\") ?? undefined;\n return json(await listTasks({ workspace, status }));\n }\n\n // POST /api/tasks - create task\n if (pathname === \"/api/tasks\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await createTask(body), 201);\n }\n\n // PUT /api/tasks/:id - update task (e.g. toggle status)\n const taskUpdateMatch = pathname.match(/^\\/api\\/tasks\\/([^/]+)$/);\n if (taskUpdateMatch && method === \"PUT\") {\n const id = taskUpdateMatch[1];\n const body = await req.json().catch(() => ({}));\n const updated = await updateTask(id, body);\n if (!updated) return json({ error: \"Not found\" }, 404);\n return json(updated);\n }\n\n // DELETE /api/tasks/:id - delete task\n const taskDeleteMatch = pathname.match(/^\\/api\\/tasks\\/([^/]+)$/);\n if (taskDeleteMatch && method === \"DELETE\") {\n const ok = await deleteTask(taskDeleteMatch[1]);\n return ok ? json({ success: true }) : json({ error: \"Not found\" }, 404);\n }\n\n // --- CRM Deals (sales pipeline) ------------------------------------------\n if (pathname === \"/api/deals\" && method === \"GET\") {\n return json(await listDeals(url.searchParams.get(\"stage\") ?? undefined));\n }\n if (pathname === \"/api/deals\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await createDeal(body), 201);\n }\n // Seed deals from qualified leads (explicit; checked before the :id matcher).\n if (pathname === \"/api/deals/sync\" && method === \"POST\") {\n return json(await syncDealsFromLeads());\n }\n // Interactive import: candidate leads to review, then import the selected emails.\n if (pathname === \"/api/deals/candidates\" && method === \"GET\") {\n return json(await listCandidates());\n }\n if (pathname === \"/api/deals/import\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n return json(await importDeals(Array.isArray(body.emails) ? body.emails : []));\n }\n const dealIdMatch = pathname.match(/^\\/api\\/deals\\/([^/]+)$/);\n if (dealIdMatch && dealIdMatch[1] !== \"sync\" && method === \"GET\") {\n const d = await getDeal(dealIdMatch[1]);\n return d ? json(d) : json({ error: \"Not found\" }, 404);\n }\n if (dealIdMatch && method === \"PUT\") {\n const body = await req.json().catch(() => ({}));\n const updated = await updateDeal(dealIdMatch[1], body);\n return updated ? json(updated) : json({ error: \"Not found\" }, 404);\n }\n if (dealIdMatch && method === \"DELETE\") {\n const ok = await deleteDeal(dealIdMatch[1]);\n return ok ? json({ success: true }) : json({ error: \"Not found\" }, 404);\n }\n\n // --- Data export (CSV) ---------------------------------------------------\n if (pathname === \"/api/export/signals\" && method === \"GET\") {\n const r = await sqlite.execute(\"SELECT id, lead_email, lead_name, signal_type, signal_source, signal_strength, actioned, detected_at FROM intent_signals ORDER BY detected_at DESC\");\n const cols = r.columns;\n let csv = cols.join(\",\") + \"\\n\";\n for (const row of r.rows) {\n csv += (row as any[]).map((v) => `\"${String(v ?? \"\").replace(/\"/g, '\"\"')}\"`).join(\",\") + \"\\n\";\n }\n return new Response(csv, { headers: { \"Content-Type\": \"text/csv\", \"Content-Disposition\": 'attachment; filename=\"signals.csv\"', ...corsHeaders } });\n }\n if (pathname === \"/api/export/okrs\" && method === \"GET\") {\n const objR = await sqlite.execute(\"SELECT id, title, description, workspace, status, progress, quarter, created_at FROM objectives ORDER BY created_at DESC\");\n const krR = await sqlite.execute(\"SELECT id, objective_id, title, target_value, current_value, unit, progress, updated_at FROM key_results ORDER BY updated_at DESC\");\n let csv = \"type,\" + objR.columns.join(\",\") + \"\\n\";\n for (const row of objR.rows) {\n csv += \"objective,\" + (row as any[]).map((v) => `\"${String(v ?? \"\").replace(/\"/g, '\"\"')}\"`).join(\",\") + \"\\n\";\n }\n if (krR.rows.length > 0) {\n csv += \"\\ntype,\" + krR.columns.join(\",\") + \"\\n\";\n for (const row of krR.rows) {\n csv += \"key_result,\" + (row as any[]).map((v) => `\"${String(v ?? \"\").replace(/\"/g, '\"\"')}\"`).join(\",\") + \"\\n\";\n }\n }\n return new Response(csv, { headers: { \"Content-Type\": \"text/csv\", \"Content-Disposition\": 'attachment; filename=\"okrs.csv\"', ...corsHeaders } });\n }\n if (pathname === \"/api/export/tasks\" && method === \"GET\") {\n const r = await sqlite.execute(\"SELECT id, title, description, status, priority, workspace, due_date, created_at, updated_at FROM tasks ORDER BY created_at DESC\");\n const cols = r.columns;\n let csv = cols.join(\",\") + \"\\n\";\n for (const row of r.rows) {\n csv += (row as any[]).map((v) => `\"${String(v ?? \"\").replace(/\"/g, '\"\"')}\"`).join(\",\") + \"\\n\";\n }\n return new Response(csv, { headers: { \"Content-Type\": \"text/csv\", \"Content-Disposition\": 'attachment; filename=\"tasks.csv\"', ...corsHeaders } });\n }\n\n // --- LinkedIn OAuth flow (legacy aliases) --------------------------------\n if (pathname === \"/api/oauth/linkedin/start\" && method === \"GET\") {\n return Response.redirect(`${url.origin}/api/auth/linkedin`);\n }\n if (pathname === \"/api/oauth/linkedin/callback\" && method === \"GET\") {\n return Response.redirect(`${url.origin}/api/auth/linkedin/callback?${url.search.slice(1)}`);\n }\n\n // --- Graphify - app map for MCP / AI-native navigation -------------------\n if (pathname === \"/api/graphify\" && method === \"GET\") {\n return json(await getGraphify());\n }\n\n // --- Calendar today shortcut (TodayPanel widget) --------------------------\n if (pathname === \"/api/calendar/today\" && method === \"GET\") {\n const today = new Date().toISOString().slice(0, 10);\n return json(await getUnifiedCalendar({ from: today, to: today }));\n }\n\n // --- Digest latest alias --------------------------------------------------\n if (pathname === \"/api/digest/latest\" && method === \"GET\") {\n return json(await getLatestDigest());\n }\n\n // --- Admin purge (clear intelligence cache) -------------------------------\n if (pathname === \"/api/admin/purge\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n if (body.confirm !== \"PURGE\") return json({ error: \"Send confirm: PURGE\" }, 400);\n await sqlite.execute(\"DELETE FROM intelligence_cache\").catch(() => {});\n await sqlite.execute(\"UPDATE app_config SET value=NULL WHERE key='funnel_cache'\").catch(() => {});\n return json({ success: true, message: \"Cache purged\" });\n }\n\n // --- Remove demo/seed data ------------------------------------------------\n // The only seeded (\"demo\") rows are the starter agents. Identify them by an exact\n // match on (name, description, system_prompt) so a renamed/edited agent is never touched.\n if (pathname === \"/api/admin/seed-status\" && method === \"GET\") {\n let agents = 0;\n for (const seed of DEFAULT_AGENT_SEEDS) {\n const r = await sqlite.execute(\n \"SELECT COUNT(*) FROM agents WHERE name = ? AND description = ? AND system_prompt = ?\",\n [seed.name, seed.description, seed.system_prompt],\n ).catch(() => null);\n if (r?.rows?.length) agents += Number((r.rows[0] as any[])[0]) || 0;\n }\n return json({ ok: true, total: agents, counts: { agents } });\n }\n if (pathname === \"/api/admin/cleanup-mock\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n if (body.confirm !== \"CLEANUP\") return json({ ok: false, error: \"Send confirm: CLEANUP\" }, 400);\n let removed = 0;\n for (const seed of DEFAULT_AGENT_SEEDS) {\n const r = await sqlite.execute(\n \"DELETE FROM agents WHERE name = ? AND description = ? AND system_prompt = ?\",\n [seed.name, seed.description, seed.system_prompt],\n ).catch(() => null);\n removed += Number((r as any)?.rowsAffected ?? 0) || 0;\n }\n return json({ ok: true, total: removed });\n }\n\n // Bulk reset: wipe ALL agents (= agents + campaigns, same table) plus the related\n // run/budget/conversation rows that deleteAgent() does not cascade. Guarded by confirm.\n if (pathname === \"/api/admin/reset-agents\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n if (body.confirm !== \"CLEANUP\") return json({ ok: false, error: \"Send confirm: CLEANUP\" }, 400);\n const counts: Record = {};\n for (const table of [\"agents\", \"agent_runs\", \"agent_budgets\", \"conversations\"]) {\n const r = await sqlite.execute(`DELETE FROM ${table}`).catch(() => null);\n counts[table] = Number((r as any)?.rowsAffected ?? 0) || 0;\n }\n return json({ ok: true, counts });\n }\n\n // Factory reset for THIS app only: wipe this app's own data tables, keeping api\n // (integrations, llm_config) + knowledge (knowledge_files). IMPORTANT: val.town's\n // std/sqlite is a single ACCOUNT-WIDE database shared by all of this user's vals, so we\n // must NEVER blanket-delete every table — only the explicit list below (the tables this\n // app creates in db.ts). Guarded by confirm.\n if (pathname === \"/api/admin/reset-data\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n if (body.confirm !== \"CLEANUP\") return json({ ok: false, error: \"Send confirm: CLEANUP\" }, 400);\n const WIPE_TABLES = [\n \"agents\", \"conversations\", \"agent_budgets\", \"agent_runs\",\n \"objectives\", \"key_results\", \"north_star\", \"nsm_history\",\n \"experiments\", \"tasks\", \"tasks_legacy\", \"notes\",\n \"intent_signals\", \"growth_digests\", \"calendar_entries\", \"action_queue\",\n \"customer_cohorts\", \"lead_state_changes\", \"utm_conversions\",\n \"metric_baselines\", \"metric_history\", \"intelligence_cache\", \"app_config\",\n ];\n const counts: Record = {};\n for (const table of WIPE_TABLES) {\n const r = await sqlite.execute(`DELETE FROM ${table}`).catch(() => null);\n counts[table] = Number((r as any)?.rowsAffected ?? 0) || 0;\n }\n return json({ ok: true, kept: [\"integrations\", \"llm_config\", \"knowledge_files\"], counts });\n }\n\n // --- Google OAuth redirect flow ------------------------------------------\n if (pathname === \"/api/auth/google\" && method === \"GET\") {\n const creds = await getCredentials(\"google\");\n // OAuth client is env-driven by design (GOOGLE_CLIENT_ID/SECRET); fall back to the DB row.\n const clientId = Deno.env.get(\"GOOGLE_CLIENT_ID\") || creds?.extra_config?.client_id || creds?.extra_config?.clientId;\n const redirectUri = creds?.extra_config?.redirect_uri || creds?.extra_config?.redirectUri || `${url.origin}/api/auth/google/callback`;\n if (!clientId) {\n return new Response(\"Google OAuth not configured. Add Client ID in Settings - Google.\", { status: 400 });\n }\n const scopes = [\n \"https://www.googleapis.com/auth/userinfo.email\",\n \"https://www.googleapis.com/auth/analytics.readonly\",\n \"https://www.googleapis.com/auth/webmasters.readonly\",\n \"https://www.googleapis.com/auth/gmail.modify\",\n \"https://www.googleapis.com/auth/drive.readonly\",\n \"https://www.googleapis.com/auth/spreadsheets.readonly\",\n \"https://www.googleapis.com/auth/documents.readonly\",\n \"https://www.googleapis.com/auth/calendar\",\n \"https://www.googleapis.com/auth/youtube.readonly\",\n ].join(\" \");\n const { state, setCookie } = newOAuthState();\n const params = new URLSearchParams({\n client_id: clientId,\n redirect_uri: redirectUri,\n response_type: \"code\",\n scope: scopes,\n access_type: \"offline\",\n prompt: \"consent\",\n state,\n });\n return redirectWithCookie(`https://accounts.google.com/o/oauth2/v2/auth?${params}`, setCookie);\n }\n\n if (pathname === \"/api/auth/google/callback\" && method === \"GET\") {\n const code = url.searchParams.get(\"code\");\n const error = url.searchParams.get(\"error\");\n if (error || !code) {\n return Response.redirect(`${url.origin}/?auth_error=${encodeURIComponent(error || \"no_code\")}`);\n }\n if (!oauthStateValid(req, url)) {\n return Response.redirect(`${url.origin}/?auth_error=state_mismatch`);\n }\n const creds = await getCredentials(\"google\");\n // Trim to avoid whitespace from copy-paste/env producing a malformed (invalid_request) token request.\n const clientId = (Deno.env.get(\"GOOGLE_CLIENT_ID\") || creds?.extra_config?.client_id || creds?.extra_config?.clientId || \"\").trim();\n const clientSecret = (Deno.env.get(\"GOOGLE_CLIENT_SECRET\") || creds?.api_secret || \"\").trim();\n const redirectUri = (creds?.extra_config?.redirect_uri || creds?.extra_config?.redirectUri || `${url.origin}/api/auth/google/callback`).trim();\n\n // Google requires non-empty client_id, client_secret, and redirect_uri for a Web-application\n // authorization_code exchange. Sending an empty required param yields error=invalid_request\n // (RFC 6749 5.2). Fail fast with a clear cause instead of a malformed request to Google.\n if (!clientId || !clientSecret) {\n const missing = !clientId && !clientSecret ? \"client_id_and_secret\"\n : !clientId ? \"client_id\" : \"client_secret\";\n return Response.redirect(`${url.origin}/?auth_error=${encodeURIComponent(\"missing_\" + missing)}`);\n }\n\n const tokenRes = await fetch(\"https://oauth2.googleapis.com/token\", {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/x-www-form-urlencoded\" },\n body: new URLSearchParams({ code, client_id: clientId, client_secret: clientSecret, redirect_uri: redirectUri, grant_type: \"authorization_code\" }),\n });\n const tokens = await tokenRes.json() as any;\n if (!tokenRes.ok || !tokens.access_token) {\n // Surface Google's error + description so invalid_request causes are diagnosable from the URL.\n const detail = [tokens.error, tokens.error_description].filter(Boolean).join(\": \") || \"token_exchange_failed\";\n return Response.redirect(`${url.origin}/?auth_error=${encodeURIComponent(detail)}`);\n }\n\n const now = new Date().toISOString();\n const newExtraConfig = { ...(creds?.extra_config || {}), token_expiry: Date.now() + (tokens.expires_in || 3600) * 1000 };\n const existing = await sqlite.execute(\"SELECT id FROM integrations WHERE name = ?\", [\"google\"]);\n if (existing.rows.length > 0) {\n await sqlite.execute(\n \"UPDATE integrations SET access_token=?, refresh_token=?, extra_config=?, updated_at=? WHERE name='google'\",\n [tokens.access_token, tokens.refresh_token || creds?.refresh_token || null, JSON.stringify(newExtraConfig), now]\n );\n } else {\n await sqlite.execute(\n \"INSERT INTO integrations (id, name, access_token, refresh_token, extra_config, connected_at, updated_at) VALUES (?,?,?,?,?,?,?)\",\n [crypto.randomUUID(), \"google\", tokens.access_token, tokens.refresh_token || null, JSON.stringify(newExtraConfig), now, now]\n );\n }\n return redirectWithCookie(`${url.origin}/?auth=google_ok`, CLEAR_OAUTH_STATE);\n }\n\n // --- LinkedIn OAuth flow --------------------------------------------------\n if (pathname === \"/api/auth/linkedin\" && method === \"GET\") {\n const creds = await getCredentials(\"linkedin\");\n const clientId = creds?.extra_config?.clientId || creds?.extra_config?.client_id;\n const redirectUri = `${url.origin}/api/auth/linkedin/callback`;\n if (!clientId) {\n return new Response(\"LinkedIn OAuth not configured. Add Client ID in Settings - LinkedIn.\", { status: 400 });\n }\n const { state, setCookie } = newOAuthState();\n const params = new URLSearchParams({\n response_type: \"code\",\n client_id: clientId,\n redirect_uri: redirectUri,\n scope: \"openid profile email w_member_social r_basicprofile\",\n state,\n });\n return redirectWithCookie(`https://www.linkedin.com/oauth/v2/authorization?${params}`, setCookie);\n }\n\n if (pathname === \"/api/auth/linkedin/callback\" && method === \"GET\") {\n const code = url.searchParams.get(\"code\");\n const error = url.searchParams.get(\"error\");\n if (error || !code) {\n return Response.redirect(`${url.origin}/?auth_error=${encodeURIComponent(error || \"no_code\")}`);\n }\n if (!oauthStateValid(req, url)) {\n return Response.redirect(`${url.origin}/?auth_error=state_mismatch`);\n }\n const creds = await getCredentials(\"linkedin\");\n const clientId = creds?.extra_config?.clientId || creds?.extra_config?.client_id || \"\";\n const clientSecret = creds?.api_secret || \"\";\n const redirectUri = `${url.origin}/api/auth/linkedin/callback`;\n\n const tokenRes = await fetch(\"https://www.linkedin.com/oauth/v2/accessToken\", {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/x-www-form-urlencoded\" },\n body: new URLSearchParams({ grant_type: \"authorization_code\", code, redirect_uri: redirectUri, client_id: clientId, client_secret: clientSecret }),\n });\n const tokens = await tokenRes.json() as any;\n if (!tokenRes.ok || !tokens.access_token) {\n return Response.redirect(`${url.origin}/?auth_error=${encodeURIComponent(tokens.error || \"linkedin_token_failed\")}`);\n }\n\n // Fetch LinkedIn profile to get the author URN needed for posts endpoint.\n let authorId = creds?.extra_config?.author_id || \"\";\n try {\n const profileRes = await fetch(\"https://api.linkedin.com/v2/me\", {\n headers: { Authorization: `Bearer ${tokens.access_token}` },\n });\n if (profileRes.ok) {\n const profile = await profileRes.json() as any;\n if (profile.id) authorId = `urn:li:person:${profile.id}`;\n }\n } catch { /* non-fatal - author_id can be set manually later */ }\n\n const now = new Date().toISOString();\n const newExtraConfig = { ...(creds?.extra_config || {}), token_expiry: Date.now() + (tokens.expires_in || 5184000) * 1000, author_id: authorId };\n const existing = await sqlite.execute(\"SELECT id FROM integrations WHERE name = ?\", [\"linkedin\"]);\n if (existing.rows.length > 0) {\n await sqlite.execute(\n \"UPDATE integrations SET access_token=?, extra_config=?, updated_at=? WHERE name='linkedin'\",\n [tokens.access_token, JSON.stringify(newExtraConfig), now]\n );\n } else {\n await sqlite.execute(\n \"INSERT INTO integrations (id, name, access_token, extra_config, connected_at, updated_at) VALUES (?,?,?,?,?,?)\",\n [crypto.randomUUID(), \"linkedin\", tokens.access_token, JSON.stringify(newExtraConfig), now, now]\n );\n }\n return redirectWithCookie(`${url.origin}/?auth=linkedin_ok`, CLEAR_OAUTH_STATE);\n }\n\n // --- Microsoft (Outlook) OAuth flow --------------------------------------\n // Env-driven client (MICROSOFT_CLIENT_ID/SECRET), `common` tenant. Mirrors the Google flow.\n if (pathname === \"/api/auth/microsoft\" && method === \"GET\") {\n // Client id/secret are env-first (MICROSOFT_CLIENT_ID/SECRET), falling back to the\n // `outlook` integration row (extra_config.client_id + api_secret) — same pattern as LinkedIn.\n const mcreds = await getCredentials(\"outlook\");\n const clientId = (Deno.env.get(\"MICROSOFT_CLIENT_ID\") || mcreds?.extra_config?.client_id || mcreds?.extra_config?.clientId || \"\").trim();\n if (!clientId) {\n return new Response(\"Microsoft OAuth not configured. Set MICROSOFT_CLIENT_ID (env) or add the Client ID in Settings → Outlook.\", { status: 400 });\n }\n const redirectUri = `${url.origin}/api/auth/microsoft/callback`;\n const scopes = MICROSOFT_SCOPES;\n const { state, setCookie } = newOAuthState();\n const params = new URLSearchParams({\n client_id: clientId,\n response_type: \"code\",\n redirect_uri: redirectUri,\n response_mode: \"query\",\n scope: scopes,\n state,\n });\n return redirectWithCookie(`https://login.microsoftonline.com/common/oauth2/v2.0/authorize?${params}`, setCookie);\n }\n\n if (pathname === \"/api/auth/microsoft/callback\" && method === \"GET\") {\n const code = url.searchParams.get(\"code\");\n const error = url.searchParams.get(\"error\");\n if (error || !code) {\n return Response.redirect(`${url.origin}/?auth_error=${encodeURIComponent(error || \"no_code\")}`);\n }\n if (!oauthStateValid(req, url)) {\n return Response.redirect(`${url.origin}/?auth_error=state_mismatch`);\n }\n const creds = await getCredentials(\"outlook\");\n const clientId = (Deno.env.get(\"MICROSOFT_CLIENT_ID\") || creds?.extra_config?.client_id || creds?.extra_config?.clientId || \"\").trim();\n const clientSecret = (Deno.env.get(\"MICROSOFT_CLIENT_SECRET\") || creds?.api_secret || \"\").trim();\n const redirectUri = `${url.origin}/api/auth/microsoft/callback`;\n if (!clientId || !clientSecret) {\n const missing = !clientId && !clientSecret ? \"client_id_and_secret\" : !clientId ? \"client_id\" : \"client_secret\";\n return Response.redirect(`${url.origin}/?auth_error=${encodeURIComponent(\"missing_microsoft_\" + missing)}`);\n }\n const scopes = MICROSOFT_SCOPES;\n const tokenRes = await fetch(\"https://login.microsoftonline.com/common/oauth2/v2.0/token\", {\n method: \"POST\",\n headers: { \"Content-Type\": \"application/x-www-form-urlencoded\" },\n body: new URLSearchParams({ client_id: clientId, client_secret: clientSecret, code, redirect_uri: redirectUri, grant_type: \"authorization_code\", scope: scopes }),\n });\n const tokens = await tokenRes.json() as any;\n if (!tokenRes.ok || !tokens.access_token) {\n const detail = [tokens.error, tokens.error_description].filter(Boolean).join(\": \") || \"token_exchange_failed\";\n return Response.redirect(`${url.origin}/?auth_error=${encodeURIComponent(detail)}`);\n }\n\n const now = new Date().toISOString();\n const newExtraConfig = { ...(creds?.extra_config || {}), token_expiry: Date.now() + (tokens.expires_in || 3600) * 1000 };\n const existing = await sqlite.execute(\"SELECT id FROM integrations WHERE name = ?\", [\"outlook\"]);\n if (existing.rows.length > 0) {\n await sqlite.execute(\n \"UPDATE integrations SET access_token=?, refresh_token=?, extra_config=?, updated_at=? WHERE name='outlook'\",\n [tokens.access_token, tokens.refresh_token || creds?.refresh_token || null, JSON.stringify(newExtraConfig), now]\n );\n } else {\n await sqlite.execute(\n \"INSERT INTO integrations (id, name, access_token, refresh_token, extra_config, connected_at, updated_at) VALUES (?,?,?,?,?,?,?)\",\n [crypto.randomUUID(), \"outlook\", tokens.access_token, tokens.refresh_token || null, JSON.stringify(newExtraConfig), now, now]\n );\n }\n return redirectWithCookie(`${url.origin}/?auth=outlook_ok`, CLEAR_OAUTH_STATE);\n }\n\n // --- Notes routes (TeamHub collaborative notes) ---------------------------\n if (pathname === \"/api/notes\" && method === \"GET\") {\n const workspace = url.searchParams.get(\"workspace\") ?? \"marketing\";\n const result = await sqlite.execute(\n \"SELECT * FROM notes WHERE workspace=? ORDER BY pinned DESC, created_at DESC\",\n [workspace]\n );\n return json(result.rows.map((r) => rowToObject(result.columns, r)));\n }\n if (pathname === \"/api/notes\" && method === \"POST\") {\n const body = await req.json().catch(() => ({}));\n const id = crypto.randomUUID();\n const now = new Date().toISOString();\n await sqlite.execute(\n \"INSERT INTO notes (id, content, pinned, workspace, created_at, updated_at) VALUES (?,?,?,?,?,?)\",\n [id, body.content ?? \"\", body.pinned ? 1 : 0, body.workspace ?? \"marketing\", now, now]\n );\n return json({ id, content: body.content ?? \"\", pinned: !!body.pinned, workspace: body.workspace ?? \"marketing\", created_at: now, updated_at: now }, 201);\n }\n const noteMatch = pathname.match(/^\\/api\\/notes\\/([^/]+)$/);\n if (noteMatch && method === \"PUT\") {\n const id = noteMatch[1];\n const body = await req.json().catch(() => ({}));\n const now = new Date().toISOString();\n const setParts: string[] = [\"updated_at=?\"];\n const setArgs: any[] = [now];\n if (body.pinned !== undefined) { setParts.push(\"pinned=?\"); setArgs.push(body.pinned ? 1 : 0); }\n if (body.content !== undefined) { setParts.push(\"content=?\"); setArgs.push(body.content); }\n await sqlite.execute(\n `UPDATE notes SET ${setParts.join(\", \")} WHERE id=?`,\n [...setArgs, id]\n );\n const r = await sqlite.execute(\"SELECT * FROM notes WHERE id=?\", [id]);\n if (!r.rows.length) return json({ error: \"Not found\" }, 404);\n return json(rowToObject(r.columns, r.rows[0]));\n }\n if (noteMatch && method === \"DELETE\") {\n await sqlite.execute(\"DELETE FROM notes WHERE id=?\", [noteMatch[1]]);\n return json({ success: true });\n }\n\n // No route matched\n return json({ error: \"Not found\", path: pathname }, 404);\n } catch (err) {\n const message = err instanceof Error ? err.message : String(err);\n console.error(`[router] Error handling ${method} ${pathname}:`, message);\n return json({ error: \"Internal server error\", detail: message }, 500);\n }\n}\n\n/**\n * Converts a sqlite result row (array of values) to a plain object\n * using the columns array from the result set.\n */\nfunction rowToObject(columns: string[], row: unknown[]): Record {\n const obj: Record = {};\n for (let i = 0; i < columns.length; i++) {\n obj[columns[i]] = row[i];\n }\n return obj;\n}\n",[176],"9d9db260-5d6b-11f1-bf4a-1607ee4eb77e","main.ts",642,["D",1780285221175],13,"http","import { initDB } from \"./db.ts\";\nimport { handleRequest } from \"./router.ts\";\nimport { getFrontendHTML } from \"./frontend.ts\";\nimport { seedKnowledge } from \"./api/knowledge_seed.ts\";\n\nlet initialized = false;\n\n// Shared-secret gate. Set APP_TOKEN in the val's environment. Every /api/* route\n// requires `Authorization: Bearer ` (or `?token=` for download links),\n// EXCEPT the public paths below. Fail-closed: if APP_TOKEN is unset, data routes 503.\nconst APP_TOKEN = (Deno.env.get(\"APP_TOKEN\") || \"\").trim();\n\n// Login verification temporarily disabled — flip to `false` to re-enable the gate.\nconst AUTH_DISABLED = true;\n\n// Public /api paths that must stay reachable without the app token:\n// - /api/app-* : the frontend's own JS modules (static code, no secrets; loaded by \n \n \n \n \n \n \n \n \n\n\n

\n \n\n`;\n}\n",[329],"9c936de2-5d6b-11f1-a71c-1607ee4eb77e","db.ts","2f68e1b8-56ca-11f1-b821-ee650bb23af1",640,["D",1780285219474],36,"import { sqlite as _sqlite } from \"https://esm.town/v/std/sqlite\";\n\n// Compatibility shim: support sqlite.execute(sql, args) two-arg form\n// Val.town's libSQL client only accepts execute(string) or execute({sql, args})\nconst _exec = _sqlite.execute.bind(_sqlite);\nexport const sqlite = {\n ..._sqlite,\n execute: async (stmt: string | { sql: string; args?: any[] }, args?: any[]): Promise => {\n if (typeof stmt === \"string\" && args !== undefined) {\n return _exec({ sql: stmt, args });\n }\n return _exec(stmt as any);\n },\n};\n\nexport async function initDB() {\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS integrations (\n id TEXT PRIMARY KEY,\n name TEXT NOT NULL UNIQUE,\n api_key TEXT,\n api_secret TEXT,\n access_token TEXT,\n refresh_token TEXT,\n extra_config TEXT,\n connected_at TEXT,\n updated_at TEXT\n )\n `);\n\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS llm_config (\n id TEXT PRIMARY KEY DEFAULT 'default',\n provider TEXT NOT NULL DEFAULT 'anthropic',\n api_key TEXT NOT NULL DEFAULT '',\n endpoint TEXT,\n model TEXT NOT NULL DEFAULT 'claude-sonnet-4-6',\n updated_at TEXT\n )\n `);\n\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS agents (\n id TEXT PRIMARY KEY,\n name TEXT NOT NULL,\n workspace TEXT NOT NULL,\n description TEXT,\n system_prompt TEXT,\n tools TEXT DEFAULT '[]',\n model TEXT,\n enabled INTEGER DEFAULT 1,\n created_at TEXT NOT NULL,\n updated_at TEXT NOT NULL,\n created_by TEXT DEFAULT 'user'\n )\n `);\n\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS conversations (\n id TEXT PRIMARY KEY,\n agent_id TEXT,\n workspace TEXT NOT NULL,\n messages TEXT DEFAULT '[]',\n created_at TEXT NOT NULL,\n updated_at TEXT NOT NULL\n )\n `);\n\n // F4 migration: the original tasks table (TEXT id, agent_id, input/output) is replaced by\n // the framework-layer tasks schema (INTEGER id, ICE fields, OKR/experiment links). If a\n // legacy table exists, rename it so the new CREATE further down can build the right shape.\n try {\n const cols = await sqlite.execute(\"PRAGMA table_info(tasks)\");\n const have = new Set((cols.rows || []).map((r: any) => String(Array.isArray(r) ? r[1] : r.name)));\n // Old schema is identifiable by `agent_id` + `input` + no `priority`. New schema has `priority` + `ice_score`.\n const isLegacy = have.has(\"agent_id\") && have.has(\"input\") && !have.has(\"ice_score\");\n if (isLegacy) {\n await sqlite.execute(\"ALTER TABLE tasks RENAME TO tasks_legacy\");\n }\n } catch { /* table doesn't exist yet — fine, new CREATE will run */ }\n\n // Co-working: shared notes for team collaboration on strategies.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS notes (\n id TEXT PRIMARY KEY,\n author TEXT,\n content TEXT NOT NULL,\n workspace TEXT,\n pinned INTEGER DEFAULT 0,\n created_at TEXT NOT NULL,\n updated_at TEXT\n )\n `);\n\n // Intelligence layer: per-metric 30d baselines used for health scoring.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS metric_baselines (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n integration TEXT NOT NULL,\n metric_key TEXT NOT NULL,\n baseline_value REAL,\n baseline_period TEXT DEFAULT '30d',\n last_updated TEXT DEFAULT (datetime('now')),\n UNIQUE(integration, metric_key)\n )\n `);\n\n // Intelligence cache — funnel summary, correlations, recommendations, LLM-generated feed.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS intelligence_cache (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n cache_key TEXT UNIQUE NOT NULL,\n payload TEXT NOT NULL,\n generated_at TEXT DEFAULT (datetime('now')),\n expires_at TEXT NOT NULL\n )\n `);\n\n // App-wide key/value config: ROI inputs (avg_deal_size, est_marketing_spend),\n // workspace preferences (ultra_density, translucency, hardware_accel, etc.).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS app_config (\n key TEXT PRIMARY KEY,\n value TEXT,\n updated_at TEXT DEFAULT (datetime('now'))\n )\n `);\n\n // Metric history — rolling timeseries for delta_7d / delta_30d computations.\n // Pruned to last 60 days on insert to keep size bounded.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS metric_history (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n integration TEXT NOT NULL,\n metric_key TEXT NOT NULL,\n value REAL NOT NULL,\n recorded_at TEXT DEFAULT (datetime('now'))\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_metric_history_key ON metric_history(integration, metric_key, recorded_at DESC)`);\n\n // A5: rolling baseline columns — added on existing metric_baselines tables via\n // PRAGMA-guarded ALTER so we never fail on re-run.\n try {\n const cols = await sqlite.execute(\"PRAGMA table_info(metric_baselines)\");\n const have = new Set((cols.rows || []).map((r: any) => String(Array.isArray(r) ? r[1] : r.name)));\n if (!have.has(\"rolling_avg\")) {\n await sqlite.execute(\"ALTER TABLE metric_baselines ADD COLUMN rolling_avg REAL\");\n }\n if (!have.has(\"sample_count\")) {\n await sqlite.execute(\"ALTER TABLE metric_baselines ADD COLUMN sample_count INTEGER DEFAULT 0\");\n }\n } catch { /* non-critical: warm-start baselines still work */ }\n\n // QA fix: per-provider key columns on llm_config so switching providers doesn't\n // overwrite the previous provider's key. PRAGMA-guarded ALTER (re-run safe).\n try {\n const cols = await sqlite.execute(\"PRAGMA table_info(llm_config)\");\n const have = new Set((cols.rows || []).map((r: any) => String(Array.isArray(r) ? r[1] : r.name)));\n for (const col of [\"api_key_anthropic\", \"api_key_openai\", \"api_key_gemini\", \"api_key_custom\"]) {\n if (!have.has(col)) await sqlite.execute(`ALTER TABLE llm_config ADD COLUMN ${col} TEXT DEFAULT ''`);\n }\n if (!have.has(\"endpoint_custom\")) {\n await sqlite.execute(\"ALTER TABLE llm_config ADD COLUMN endpoint_custom TEXT DEFAULT ''\");\n }\n } catch { /* non-critical: legacy single-key still works */ }\n\n // Agent budget tracking — real monthly LLM-cost cap per agent.\n // current_spend incremented on each LLM call, reset at period_start rollover.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS agent_budgets (\n agent_id TEXT PRIMARY KEY,\n monthly_cap REAL DEFAULT 250,\n current_spend REAL DEFAULT 0,\n period_start TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now'))\n )\n `);\n\n // C3: agent execution log. One row per run (manual or scheduled). Latest 50 power the\n // AgentCard timeline. status: queued → running → completed | failed | cancelled.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS agent_runs (\n id TEXT PRIMARY KEY,\n agent_id TEXT NOT NULL,\n status TEXT NOT NULL,\n trigger TEXT DEFAULT 'manual',\n started_at TEXT DEFAULT (datetime('now')),\n finished_at TEXT,\n duration_ms INTEGER,\n input TEXT,\n output TEXT,\n error TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_agent_runs_agent ON agent_runs(agent_id, started_at DESC)`);\n\n // D1: customer cohorts. One row per cohort (e.g. signup month). Powers cohort-survival LTV\n // beyond the simple avg_deal × retention_months heuristic. retention_curve is a JSON array\n // [m1, m2, m3 ...] of retained customer counts. arpu = average revenue per user (monthly).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS customer_cohorts (\n id TEXT PRIMARY KEY,\n cohort_label TEXT NOT NULL,\n cohort_size INTEGER NOT NULL,\n arpu REAL NOT NULL,\n retention_curve TEXT NOT NULL,\n created_at TEXT DEFAULT (datetime('now'))\n )\n `);\n\n // D3: lead state changes — every MQL→SQL transition gets a row so we can compute\n // engagement velocity (avg days lead spent in each stage).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS lead_state_changes (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n lead_email TEXT NOT NULL,\n from_state TEXT,\n to_state TEXT NOT NULL,\n changed_at TEXT DEFAULT (datetime('now'))\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_lead_state_email ON lead_state_changes(lead_email, changed_at)`);\n\n // D5: utm_conversions — channel attribution log. Populated when a conversion fires\n // (manually pushed via POST /api/attribution or scraped from GA4 in a later phase).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS utm_conversions (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n utm_source TEXT,\n utm_medium TEXT,\n utm_campaign TEXT,\n conversion_value REAL DEFAULT 0,\n converted_at TEXT DEFAULT (datetime('now'))\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_utm_source ON utm_conversions(utm_source, converted_at DESC)`);\n\n // ─── FRAMEWORK LAYER (F1-F4): NSM, OKRs, Experiments, Tasks ─────────────────\n\n // F1: North Star Metric (singleton — only one active row; previous deleted on set)\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS north_star (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n metric_name TEXT NOT NULL,\n metric_source TEXT NOT NULL,\n metric_key TEXT,\n current_value REAL,\n target_value REAL NOT NULL,\n target_date TEXT NOT NULL,\n unit TEXT DEFAULT 'count',\n description TEXT,\n set_at TEXT DEFAULT (datetime('now')),\n set_by TEXT DEFAULT 'admin'\n )\n `);\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS nsm_history (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n nsm_id INTEGER REFERENCES north_star(id) ON DELETE CASCADE,\n value REAL NOT NULL,\n recorded_at TEXT DEFAULT (datetime('now'))\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_nsm_history_id ON nsm_history(nsm_id, recorded_at DESC)`);\n\n // F2: OKRs (Objectives + Key Results)\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS objectives (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n title TEXT NOT NULL,\n description TEXT,\n quarter TEXT NOT NULL,\n workspace TEXT DEFAULT 'marketing',\n owner TEXT,\n status TEXT DEFAULT 'active',\n progress REAL DEFAULT 0,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now'))\n )\n `);\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS key_results (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n objective_id INTEGER NOT NULL REFERENCES objectives(id) ON DELETE CASCADE,\n title TEXT NOT NULL,\n metric_source TEXT,\n metric_key TEXT,\n start_value REAL DEFAULT 0,\n current_value REAL DEFAULT 0,\n target_value REAL NOT NULL,\n unit TEXT DEFAULT 'count',\n confidence REAL DEFAULT 0.5,\n progress REAL DEFAULT 0,\n updated_at TEXT DEFAULT (datetime('now'))\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_kr_objective ON key_results(objective_id)`);\n\n // F3: ICE Experiments (ice_score + uplift_pct computed in app code — sqlite GENERATED\n // column support is uneven; safer to project them client-side from impact/confidence/ease).\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS experiments (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n title TEXT NOT NULL,\n hypothesis TEXT,\n objective_id INTEGER REFERENCES objectives(id) ON DELETE SET NULL,\n funnel_stage TEXT,\n aarrr_lever TEXT,\n status TEXT DEFAULT 'idea',\n impact REAL DEFAULT 5,\n confidence REAL DEFAULT 5,\n ease REAL DEFAULT 5,\n owner TEXT,\n start_date TEXT,\n end_date TEXT,\n channel TEXT,\n success_metric TEXT,\n baseline_value REAL,\n result_value REAL,\n outcome TEXT,\n learnings TEXT,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now'))\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_exp_status ON experiments(status, updated_at DESC)`);\n\n // F4: Tasks (linked to OKRs / KRs / experiments — at least one for governance).\n // NOTE: the gap-closure prompt also referred to a \"tasks\" table — the older one (string id,\n // status enum 'pending|running|done|failed') is replaced by this richer schema. The two\n // were never wired to a shared UI so the swap is safe.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS tasks (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n title TEXT NOT NULL,\n description TEXT,\n status TEXT DEFAULT 'todo',\n priority TEXT DEFAULT 'medium',\n workspace TEXT DEFAULT 'marketing',\n assignee TEXT,\n due_date TEXT,\n objective_id INTEGER REFERENCES objectives(id) ON DELETE SET NULL,\n key_result_id INTEGER REFERENCES key_results(id) ON DELETE SET NULL,\n experiment_id INTEGER REFERENCES experiments(id) ON DELETE SET NULL,\n ice_score REAL,\n impact REAL,\n confidence REAL,\n ease REAL,\n tags TEXT,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now')),\n completed_at TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_tasks_status ON tasks(status, due_date)`);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_tasks_links ON tasks(experiment_id, key_result_id, objective_id)`);\n\n // ─── INTELLIGENCE LAYER EXTENSIONS (I1, I4): Signals + Digests ──────────────\n\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS intent_signals (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n lead_email TEXT NOT NULL,\n lead_name TEXT,\n company TEXT,\n signal_type TEXT NOT NULL,\n signal_source TEXT NOT NULL,\n signal_strength INTEGER DEFAULT 1,\n raw_data TEXT,\n detected_at TEXT DEFAULT (datetime('now')),\n actioned INTEGER DEFAULT 0,\n actioned_at TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_signals_email ON intent_signals(lead_email, detected_at DESC)`);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_signals_actioned ON intent_signals(actioned, detected_at DESC)`);\n\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS growth_digests (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n week_of TEXT NOT NULL,\n payload TEXT NOT NULL,\n nsm_value_at_generation REAL,\n generated_at TEXT DEFAULT (datetime('now'))\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_digest_week ON growth_digests(week_of DESC)`);\n\n // ─── UNIFIED CALENDAR (CAL1): standalone custom calendar entries ────────────\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS calendar_entries (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n title TEXT NOT NULL,\n description TEXT,\n event_date TEXT NOT NULL,\n event_time TEXT,\n end_date TEXT,\n end_time TEXT,\n event_type TEXT DEFAULT 'custom',\n workspace TEXT DEFAULT 'marketing',\n color TEXT DEFAULT 'blue',\n all_day INTEGER DEFAULT 0,\n recurrence TEXT,\n created_at TEXT DEFAULT (datetime('now'))\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_calendar_date ON calendar_entries(event_date)`);\n\n // Knowledge files — growth playbooks/frameworks/formulas injected into AI features (see api/knowledge.ts).\n // This table was missing from initDB, so the knowledge UI + /api/knowledge endpoints failed (\"no such table\").\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS knowledge_files (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n name TEXT NOT NULL,\n kind TEXT DEFAULT 'playbook',\n content TEXT NOT NULL,\n summary TEXT,\n targets TEXT DEFAULT '[]',\n source TEXT DEFAULT 'user',\n tokens_estimate INTEGER DEFAULT 0,\n enabled INTEGER DEFAULT 1,\n uploaded_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now'))\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_knowledge_enabled ON knowledge_files(enabled, uploaded_at DESC)`);\n\n // Action queue — human-in-the-loop execution. The AI/agents PREPARE external actions\n // (send email, publish post, post to LinkedIn, create calendar event) here; nothing is\n // sent until a human approves, at which point the integration call runs and the result\n // is recorded. status: prepared → approved → executing → done | failed | rejected.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS action_queue (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n source_type TEXT DEFAULT 'chat',\n source_id TEXT,\n integration TEXT NOT NULL,\n endpoint TEXT NOT NULL,\n payload TEXT NOT NULL,\n preview TEXT,\n title TEXT,\n workspace TEXT DEFAULT 'marketing',\n status TEXT DEFAULT 'prepared',\n result TEXT,\n task_id TEXT,\n created_at TEXT DEFAULT (datetime('now')),\n approved_at TEXT,\n executed_at TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_action_queue_status ON action_queue(status, created_at DESC)`);\n\n // CRM: sales pipeline deals. Hybrid model — rows are auto-seeded from qualified leads\n // (getLeadScores) via POST /api/deals/sync AND can be created/edited/dragged by the user.\n // stage ∈ lead | qualified | proposal | negotiation | won | lost. lead_email links a deal back\n // to its source lead so the detail view can show that contact's intent_signals as activity.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS deals (\n id TEXT PRIMARY KEY,\n title TEXT,\n company TEXT,\n contact_name TEXT,\n contact_email TEXT,\n stage TEXT NOT NULL DEFAULT 'lead',\n amount REAL DEFAULT 0,\n currency TEXT DEFAULT 'USD',\n probability INTEGER DEFAULT 0,\n owner TEXT,\n close_date TEXT,\n source TEXT DEFAULT 'manual',\n lead_email TEXT,\n notes TEXT,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now')),\n last_activity_at TEXT\n )\n `);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_deals_stage ON deals(stage, updated_at DESC)`);\n await sqlite.execute(`CREATE INDEX IF NOT EXISTS idx_deals_lead ON deals(lead_email)`);\n\n // I3: agents.role column for Listener/Topic/Creator/Analyst/General roles (ALTER guarded)\n try {\n const ac = await sqlite.execute(\"PRAGMA table_info(agents)\");\n const aHave = new Set((ac.rows || []).map((r: any) => String(Array.isArray(r) ? r[1] : r.name)));\n if (!aHave.has(\"role\")) await sqlite.execute(\"ALTER TABLE agents ADD COLUMN role TEXT DEFAULT 'general'\");\n } catch { /* non-critical */ }\n\n // Part 15.3: persistent business memory (evolving learned facts). Additive on v551.\n await sqlite.execute(`\n CREATE TABLE IF NOT EXISTS memory (\n id INTEGER PRIMARY KEY AUTOINCREMENT,\n type TEXT DEFAULT 'fact',\n content TEXT NOT NULL,\n source TEXT DEFAULT 'manual',\n confidence REAL DEFAULT 0.7,\n pinned INTEGER DEFAULT 0,\n created_at TEXT DEFAULT (datetime('now')),\n updated_at TEXT DEFAULT (datetime('now'))\n )\n `);\n\n // Seed default LLM config if not exists\n const existing = await sqlite.execute(\"SELECT id FROM llm_config WHERE id = 'default'\");\n if (existing.rows.length === 0) {\n await sqlite.execute(\n \"INSERT INTO llm_config (id, provider, api_key, model, updated_at) VALUES ('default', 'anthropic', '', 'claude-sonnet-4-6', ?)\",\n [new Date().toISOString()]\n );\n }\n\n // Demo-agent seeding disabled — starting from a clean agents table.\n // Re-enable by uncommenting the call below.\n // await seedDefaultAgents();\n}\n\n// The starter \"demo\" agents seeded on first boot. Exported so the admin \"Remove demo data\"\n// route can identify + remove untouched seed rows (matched by name+description+system_prompt,\n// so a user-renamed/edited agent is never deleted).\nexport const DEFAULT_AGENT_SEEDS = [\n {\n name: \"SEO Analyst\",\n workspace: \"marketing\",\n description: \"Analyzes SEO performance using Ahrefs and Google Search Console data\",\n system_prompt: \"You are an expert SEO analyst. Analyze search performance data, identify opportunities, and provide actionable recommendations to improve organic search rankings.\",\n tools: JSON.stringify([\"ahrefs\", \"gsc\", \"ga4\"]),\n model: \"claude-sonnet-4-6\",\n },\n {\n name: \"Content Creator\",\n workspace: \"marketing\",\n description: \"Creates and manages content for WordPress, YouTube, and social media\",\n system_prompt: \"You are a creative content strategist. Help plan, write, and optimize content for blogs, videos, and social media platforms to drive engagement and conversions.\",\n tools: JSON.stringify([\"wordpress\", \"youtube\", \"ga4\"]),\n model: \"claude-sonnet-4-6\",\n },\n {\n name: \"Analytics Reporter\",\n workspace: \"marketing\",\n description: \"Generates reports from GA4 and Search Console data\",\n system_prompt: \"You are a data analyst specializing in digital marketing analytics. Create insightful reports from GA4 and Search Console data to track KPIs and identify growth opportunities.\",\n tools: JSON.stringify([\"ga4\", \"gsc\", \"ahrefs\"]),\n model: \"claude-sonnet-4-6\",\n },\n {\n name: \"LinkedIn Outreach\",\n workspace: \"sales\",\n description: \"Manages LinkedIn prospecting and connection campaigns\",\n system_prompt: \"You are a sales development expert specializing in LinkedIn outreach. Help craft personalized connection requests, follow-up messages, and engagement strategies to build pipeline.\",\n tools: JSON.stringify([\"linkedin\", \"outlook\"]),\n model: \"claude-sonnet-4-6\",\n },\n {\n name: \"Lead Researcher\",\n workspace: \"sales\",\n description: \"Researches prospects using LinkedIn and web data\",\n system_prompt: \"You are a research specialist who finds and qualifies sales leads. Analyze LinkedIn profiles, company data, and online presence to identify ideal prospects and gather intelligence for sales outreach.\",\n tools: JSON.stringify([\"linkedin\"]),\n model: \"claude-sonnet-4-6\",\n },\n {\n name: \"Email Composer\",\n workspace: \"sales\",\n description: \"Drafts and sends personalized sales emails via Outlook\",\n system_prompt: \"You are a sales email expert who crafts compelling, personalized outreach emails. Write emails that are concise, value-focused, and drive response rates while maintaining brand voice.\",\n tools: JSON.stringify([\"outlook\", \"linkedin\"]),\n model: \"claude-sonnet-4-6\",\n },\n];\n\nasync function seedDefaultAgents() {\n const existing = await sqlite.execute(\"SELECT id FROM agents LIMIT 1\");\n if (existing.rows.length > 0) return;\n\n const now = new Date().toISOString();\n for (const agent of DEFAULT_AGENT_SEEDS) {\n await sqlite.execute(\n `INSERT INTO agents (id, name, workspace, description, system_prompt, tools, model, enabled, created_at, updated_at, created_by)\n VALUES (?, ?, ?, ?, ?, ?, ?, 1, ?, ?, ?)`,\n [crypto.randomUUID(), agent.name, agent.workspace, agent.description, agent.system_prompt, agent.tools, agent.model, now, now, \"user\"]\n );\n }\n}\n\n",[337],"fb1075ca-5b1a-11f1-b807-ee650bb23af1","cron_entry.ts","fb10767e-5b1a-11f1-b807-ee650bb23af1",476,["D",1780030686427],"// Scheduled entrypoint — deploy this file as a Val Town INTERVAL/CRON trigger (separate from\n// the HTTP val in main.ts). A single file can't be both the HTTP handler AND the scheduled\n// trigger, which is why the execution engine never ran on its own before: cron.ts had a default\n// export but nothing scheduled it.\n//\n// What it does each tick:\n// 1) processPendingRuns() — runs the LLM tool-loop for any agent_runs with status='pending'\n// (external sends still go to the human-approval action_queue; nothing auto-sends).\n// 2) syncObjectivesFromFunnel() — refreshes + persists KR/objective progress from live funnel\n// data (the GET /api/okrs list no longer writes on read, so this keeps stored progress fresh).\n//\n// Schedule: set the interval in the Val Town dashboard (Triggers → e.g. every 5 minutes).\nimport { initDB } from \"./db.ts\";\nimport { processPendingRuns } from \"./cron.ts\";\nimport { syncObjectivesFromFunnel } from \"./api/okrs.ts\";\n\nexport default async function (): Promise {\n await initDB();\n const runs = await processPendingRuns(5);\n const okr = await syncObjectivesFromFunnel();\n console.log(\n `[cron] runs processed=${runs.processed} failed=${runs.failed}; ` +\n `okr synced objectives=${okr.objectives} key_results=${okr.key_results}`,\n );\n}\n",[345],"2c0d34e2-5d69-11f1-b581-1607ee4eb77e","cron.ts","50e655da-5a69-11f1-ab8c-ee650bb23af1",624,["D",1780284171703],"// Execution engine — the missing piece that makes agents actually RUN.\n// Deploy as a val.town interval val (default export runs on a schedule, e.g. every 1–5 min),\n// or trigger manually via POST /api/agents/run-pending (wired in router.ts).\n//\n// It polls agent_runs for status='pending', loads the agent config, and runs an LLM\n// tool-loop reusing chat.ts's buildTools()/executeTool(). External sends still go to the\n// human-approval action_queue (the tools enqueue, they don't send) — so the engine is safe.\nimport { sqlite } from \"./db.ts\";\nimport { getLLMConfig, callLLM } from \"./api/llm.ts\";\nimport { getAgent } from \"./api/agents.ts\";\nimport { buildTools, executeTool } from \"./api/chat.ts\";\n\nconst ENGINE_SYSTEM_SUFFIX =\n \"\\n\\nYou can ACT using tools. External sends (email / LinkedIn / WordPress) are QUEUED for human \" +\n \"approval — prepare them and they will appear in Approvals; they will NOT send until a human approves. \" +\n \"Create tasks and schedule calendar events directly. Take the single next-best action toward the goal.\";\n\nexport async function processPendingRuns(limit = 3): Promise<{ processed: number; failed: number }> {\n let processed = 0;\n let failed = 0;\n let rows: any[] = [];\n try {\n const r = await sqlite.execute(\n \"SELECT id, agent_id, input FROM agent_runs WHERE status = 'pending' ORDER BY started_at ASC LIMIT ?\",\n [limit],\n );\n rows = (r.rows || []) as any[];\n } catch {\n return { processed, failed };\n }\n\n for (const row of rows) {\n const runId = String(row[0]);\n const agentId = String(row[1]);\n const input = row[2] ? String(row[2]) : \"\";\n await sqlite.execute(\"UPDATE agent_runs SET status='running' WHERE id=?\", [runId]).catch(() => {});\n try {\n const agent = await getAgent(agentId);\n const llmConfig = await getLLMConfig();\n if (!(llmConfig as any).api_key) throw new Error(\"No LLM API key configured\");\n\n const systemPrompt =\n (agent?.system_prompt || \"You are an autonomous marketing & sales agent.\") + ENGINE_SYSTEM_SUFFIX;\n const messages: any[] = [{\n role: \"user\",\n content: input || `Run agent \"${agent?.name || agentId}\". Decide and prepare the next best action toward its goal.`,\n }];\n\n const tools = buildTools();\n let llm = await callLLM(llmConfig, messages, { systemPrompt, tools, maxTokens: 2048 });\n const actionLog: string[] = [];\n let rounds = 0;\n while (llm.tool_calls && llm.tool_calls.length > 0 && rounds < 4) {\n rounds++;\n messages.push({ role: \"assistant\", content: llm.content || \"\", tool_calls: llm.tool_calls });\n for (const tc of llm.tool_calls) {\n let args: Record = {};\n try {\n args = typeof tc.function.arguments === \"string\"\n ? JSON.parse(tc.function.arguments)\n : (tc.function.arguments || {});\n } catch { args = {}; }\n const res = await executeTool(tc.function.name, args, llmConfig);\n actionLog.push(`${tc.function.name}: ${res}`);\n messages.push({ role: \"tool\", tool_call_id: tc.id, content: res });\n }\n llm = await callLLM(llmConfig, messages, { systemPrompt, tools, maxTokens: 2048 });\n }\n\n const output = ((llm.content || \"\") + (actionLog.length ? \"\\n\\n[actions]\\n\" + actionLog.join(\"\\n\") : \"\")).slice(0, 8000);\n await sqlite.execute(\n \"UPDATE agent_runs SET status='completed', finished_at=datetime('now'), output=? WHERE id=?\",\n [output, runId],\n );\n processed++;\n } catch (e: any) {\n failed++;\n await sqlite.execute(\n \"UPDATE agent_runs SET status='failed', finished_at=datetime('now'), error=? WHERE id=?\",\n [String(e?.message || e), runId],\n ).catch(() => {});\n }\n }\n return { processed, failed };\n}\n\n// val.town interval-val entrypoint.\nexport default async function (): Promise {\n const result = await processPendingRuns(3);\n console.log(`[cron] agent runs processed=${result.processed} failed=${result.failed}`);\n}\n",[352],"eb52f0ae-5d6d-11f1-95b0-1607ee4eb77e","skills","00cb9425-5deb-432c-b10f-0fcd3d61109e",644,["D",1780286210534],[359],"d0f43d42-5b4e-11f1-8d86-ee650bb23af1","renderers","d81e0fb0-4134-482f-8b60-863ef03f710e",520,["D",1780052949625],[365],"156bfa70-56ca-11f1-8a9a-ee650bb23af1","api","e712dcc1-c69e-4e42-a401-18299fcf7843",["D",1779556136787],[371],"pulls","branches","forks","envVars",{"_382":384,"_385":69},"connections",{"_382":283,"_383":69},"count","unfilledCount",4,"emptyCount",{"_49":370,"_32":371,"_30":372,"_167":-5,"_41":42,"_40":196,"_169":65,"_38":1120,"_13":69,"_34":47,"_36":-5,"_24":25,"_173":-5,"_487":1121,"_27":1122},{"_49":1115,"_32":1000,"_30":994,"_167":372,"_41":42,"_40":1116,"_169":65,"_38":1117,"_13":69,"_34":47,"_36":-5,"_24":25,"_173":-5,"_487":1118,"_27":1119},{"_49":1107,"_32":1108,"_30":1109,"_167":994,"_41":42,"_40":1110,"_169":25,"_38":1111,"_13":898,"_34":35,"_36":1112,"_24":25,"_173":-5,"_487":1113,"_27":1114},{"_49":1100,"_32":771,"_30":1101,"_167":994,"_41":42,"_40":1102,"_169":25,"_38":1103,"_13":69,"_34":35,"_36":1104,"_24":25,"_173":-5,"_487":1105,"_27":1106},{"_49":1093,"_32":763,"_30":1094,"_167":994,"_41":42,"_40":1095,"_169":25,"_38":1096,"_13":69,"_34":35,"_36":1097,"_24":25,"_173":-5,"_487":1098,"_27":1099},{"_49":1086,"_32":755,"_30":1087,"_167":994,"_41":42,"_40":1088,"_169":25,"_38":1089,"_13":196,"_34":35,"_36":1090,"_24":25,"_173":-5,"_487":1091,"_27":1092},{"_49":1079,"_32":747,"_30":1080,"_167":994,"_41":42,"_40":1081,"_169":25,"_38":1082,"_13":384,"_34":35,"_36":1083,"_24":25,"_173":-5,"_487":1084,"_27":1085},{"_49":1072,"_32":739,"_30":1073,"_167":994,"_41":42,"_40":1074,"_169":65,"_38":1075,"_13":69,"_34":35,"_36":1076,"_24":25,"_173":-5,"_487":1077,"_27":1078},{"_49":1065,"_32":731,"_30":1066,"_167":994,"_41":42,"_40":1067,"_169":25,"_38":1068,"_13":196,"_34":35,"_36":1069,"_24":25,"_173":-5,"_487":1070,"_27":1071},{"_49":1058,"_32":723,"_30":1059,"_167":994,"_41":42,"_40":1060,"_169":25,"_38":1061,"_13":69,"_34":35,"_36":1062,"_24":25,"_173":-5,"_487":1063,"_27":1064},{"_49":1051,"_32":715,"_30":1052,"_167":994,"_41":42,"_40":1053,"_169":25,"_38":1054,"_13":196,"_34":35,"_36":1055,"_24":25,"_173":-5,"_487":1056,"_27":1057},{"_49":1043,"_32":707,"_30":1044,"_167":994,"_41":42,"_40":1045,"_169":25,"_38":1046,"_13":1047,"_34":35,"_36":1048,"_24":25,"_173":-5,"_487":1049,"_27":1050},{"_49":1036,"_32":699,"_30":1037,"_167":994,"_41":42,"_40":1038,"_169":25,"_38":1039,"_13":384,"_34":35,"_36":1040,"_24":25,"_173":-5,"_487":1041,"_27":1042},{"_49":1029,"_32":691,"_30":1030,"_167":994,"_41":42,"_40":1031,"_169":25,"_38":1032,"_13":275,"_34":35,"_36":1033,"_24":25,"_173":-5,"_487":1034,"_27":1035},{"_49":1022,"_32":683,"_30":1023,"_167":994,"_41":42,"_40":1024,"_169":65,"_38":1025,"_13":69,"_34":35,"_36":1026,"_24":25,"_173":-5,"_487":1027,"_27":1028},{"_49":1015,"_32":675,"_30":1016,"_167":994,"_41":42,"_40":1017,"_169":65,"_38":1018,"_13":69,"_34":35,"_36":1019,"_24":25,"_173":-5,"_487":1020,"_27":1021},{"_49":1008,"_32":667,"_30":1009,"_167":994,"_41":42,"_40":1010,"_169":65,"_38":1011,"_13":69,"_34":35,"_36":1012,"_24":25,"_173":-5,"_487":1013,"_27":1014},{"_49":1001,"_32":659,"_30":1002,"_167":994,"_41":42,"_40":1003,"_169":25,"_38":1004,"_13":196,"_34":35,"_36":1005,"_24":25,"_173":-5,"_487":1006,"_27":1007},{"_49":992,"_32":651,"_30":993,"_167":994,"_41":42,"_40":995,"_169":25,"_38":996,"_13":69,"_34":35,"_36":997,"_24":25,"_173":-5,"_487":998,"_27":999},{"_49":984,"_32":985,"_30":986,"_167":372,"_41":42,"_40":987,"_169":25,"_38":988,"_13":69,"_34":35,"_36":989,"_24":25,"_173":-5,"_487":990,"_27":991},{"_49":976,"_32":977,"_30":978,"_167":372,"_41":42,"_40":979,"_169":25,"_38":980,"_13":171,"_34":35,"_36":981,"_24":25,"_173":-5,"_487":982,"_27":983},{"_49":968,"_32":969,"_30":970,"_167":372,"_41":42,"_40":971,"_169":25,"_38":972,"_13":275,"_34":35,"_36":973,"_24":25,"_173":-5,"_487":974,"_27":975},{"_49":960,"_32":961,"_30":962,"_167":372,"_41":42,"_40":963,"_169":25,"_38":964,"_13":384,"_34":35,"_36":965,"_24":25,"_173":-5,"_487":966,"_27":967},{"_49":951,"_32":952,"_30":953,"_167":372,"_41":42,"_40":954,"_169":25,"_38":955,"_13":956,"_34":35,"_36":957,"_24":25,"_173":-5,"_487":958,"_27":959},{"_49":943,"_32":944,"_30":945,"_167":372,"_41":42,"_40":946,"_169":25,"_38":947,"_13":69,"_34":35,"_36":948,"_24":25,"_173":-5,"_487":949,"_27":950},{"_49":935,"_32":936,"_30":937,"_167":372,"_41":42,"_40":938,"_169":25,"_38":939,"_13":220,"_34":35,"_36":940,"_24":25,"_173":-5,"_487":941,"_27":942},{"_49":927,"_32":928,"_30":929,"_167":372,"_41":42,"_40":930,"_169":25,"_38":931,"_13":196,"_34":35,"_36":932,"_24":25,"_173":-5,"_487":933,"_27":934},{"_49":919,"_32":920,"_30":921,"_167":372,"_41":42,"_40":922,"_169":25,"_38":923,"_13":304,"_34":35,"_36":924,"_24":25,"_173":-5,"_487":925,"_27":926},{"_49":911,"_32":912,"_30":913,"_167":372,"_41":42,"_40":914,"_169":65,"_38":915,"_13":69,"_34":35,"_36":916,"_24":25,"_173":-5,"_487":917,"_27":918},{"_49":902,"_32":903,"_30":904,"_167":372,"_41":42,"_40":905,"_169":25,"_38":906,"_13":907,"_34":35,"_36":908,"_24":25,"_173":-5,"_487":909,"_27":910},{"_49":893,"_32":894,"_30":895,"_167":372,"_41":42,"_40":896,"_169":25,"_38":897,"_13":898,"_34":35,"_36":899,"_24":25,"_173":-5,"_487":900,"_27":901},{"_49":885,"_32":886,"_30":887,"_167":372,"_41":42,"_40":888,"_169":25,"_38":889,"_13":275,"_34":35,"_36":890,"_24":25,"_173":-5,"_487":891,"_27":892},{"_49":877,"_32":878,"_30":879,"_167":372,"_41":42,"_40":880,"_169":25,"_38":881,"_13":69,"_34":35,"_36":882,"_24":25,"_173":-5,"_487":883,"_27":884},{"_49":868,"_32":869,"_30":870,"_167":372,"_41":42,"_40":871,"_169":25,"_38":872,"_13":873,"_34":35,"_36":874,"_24":25,"_173":-5,"_487":875,"_27":876},{"_49":860,"_32":861,"_30":862,"_167":372,"_41":42,"_40":863,"_169":65,"_38":864,"_13":283,"_34":35,"_36":865,"_24":25,"_173":-5,"_487":866,"_27":867},{"_49":852,"_32":853,"_30":854,"_167":372,"_41":42,"_40":855,"_169":25,"_38":856,"_13":69,"_34":35,"_36":857,"_24":25,"_173":-5,"_487":858,"_27":859},{"_49":844,"_32":845,"_30":846,"_167":372,"_41":42,"_40":847,"_169":25,"_38":848,"_13":69,"_34":35,"_36":849,"_24":25,"_173":-5,"_487":850,"_27":851},{"_49":836,"_32":837,"_30":838,"_167":372,"_41":42,"_40":839,"_169":25,"_38":840,"_13":275,"_34":35,"_36":841,"_24":25,"_173":-5,"_487":842,"_27":843},{"_49":828,"_32":829,"_30":830,"_167":372,"_41":42,"_40":831,"_169":25,"_38":832,"_13":251,"_34":35,"_36":833,"_24":25,"_173":-5,"_487":834,"_27":835},{"_49":820,"_32":821,"_30":822,"_167":372,"_41":42,"_40":823,"_169":25,"_38":824,"_13":196,"_34":35,"_36":825,"_24":25,"_173":-5,"_487":826,"_27":827},{"_49":812,"_32":813,"_30":814,"_167":372,"_41":42,"_40":815,"_169":25,"_38":816,"_13":384,"_34":35,"_36":817,"_24":25,"_173":-5,"_487":818,"_27":819},{"_49":804,"_32":805,"_30":806,"_167":372,"_41":42,"_40":807,"_169":25,"_38":808,"_13":69,"_34":35,"_36":809,"_24":25,"_173":-5,"_487":810,"_27":811},{"_49":796,"_32":797,"_30":798,"_167":372,"_41":42,"_40":799,"_169":65,"_38":800,"_13":69,"_34":35,"_36":801,"_24":25,"_173":-5,"_487":802,"_27":803},{"_49":788,"_32":789,"_30":790,"_167":372,"_41":42,"_40":791,"_169":25,"_38":792,"_13":220,"_34":35,"_36":793,"_24":25,"_173":-5,"_487":794,"_27":795},{"_49":364,"_32":365,"_30":366,"_167":-5,"_41":42,"_40":367,"_169":65,"_38":785,"_13":69,"_34":47,"_36":-5,"_24":25,"_173":-5,"_487":786,"_27":787},{"_49":778,"_32":779,"_30":780,"_167":366,"_41":42,"_40":367,"_169":65,"_38":781,"_13":69,"_34":35,"_36":782,"_24":25,"_173":-5,"_487":783,"_27":784},{"_49":770,"_32":771,"_30":772,"_167":366,"_41":42,"_40":773,"_169":65,"_38":774,"_13":69,"_34":35,"_36":775,"_24":25,"_173":-5,"_487":776,"_27":777},{"_49":762,"_32":763,"_30":764,"_167":366,"_41":42,"_40":765,"_169":65,"_38":766,"_13":69,"_34":35,"_36":767,"_24":25,"_173":-5,"_487":768,"_27":769},{"_49":754,"_32":755,"_30":756,"_167":366,"_41":42,"_40":757,"_169":65,"_38":758,"_13":283,"_34":35,"_36":759,"_24":25,"_173":-5,"_487":760,"_27":761},{"_49":746,"_32":747,"_30":748,"_167":366,"_41":42,"_40":749,"_169":65,"_38":750,"_13":69,"_34":35,"_36":751,"_24":25,"_173":-5,"_487":752,"_27":753},{"_49":738,"_32":739,"_30":740,"_167":366,"_41":42,"_40":741,"_169":65,"_38":742,"_13":69,"_34":35,"_36":743,"_24":25,"_173":-5,"_487":744,"_27":745},{"_49":730,"_32":731,"_30":732,"_167":366,"_41":42,"_40":733,"_169":25,"_38":734,"_13":69,"_34":35,"_36":735,"_24":25,"_173":-5,"_487":736,"_27":737},{"_49":722,"_32":723,"_30":724,"_167":366,"_41":42,"_40":725,"_169":65,"_38":726,"_13":69,"_34":35,"_36":727,"_24":25,"_173":-5,"_487":728,"_27":729},{"_49":714,"_32":715,"_30":716,"_167":366,"_41":42,"_40":717,"_169":25,"_38":718,"_13":69,"_34":35,"_36":719,"_24":25,"_173":-5,"_487":720,"_27":721},{"_49":706,"_32":707,"_30":708,"_167":366,"_41":42,"_40":709,"_169":65,"_38":710,"_13":69,"_34":35,"_36":711,"_24":25,"_173":-5,"_487":712,"_27":713},{"_49":698,"_32":699,"_30":700,"_167":366,"_41":42,"_40":701,"_169":65,"_38":702,"_13":69,"_34":35,"_36":703,"_24":25,"_173":-5,"_487":704,"_27":705},{"_49":690,"_32":691,"_30":692,"_167":366,"_41":42,"_40":693,"_169":65,"_38":694,"_13":69,"_34":35,"_36":695,"_24":25,"_173":-5,"_487":696,"_27":697},{"_49":682,"_32":683,"_30":684,"_167":366,"_41":42,"_40":685,"_169":65,"_38":686,"_13":69,"_34":35,"_36":687,"_24":25,"_173":-5,"_487":688,"_27":689},{"_49":674,"_32":675,"_30":676,"_167":366,"_41":42,"_40":677,"_169":65,"_38":678,"_13":283,"_34":35,"_36":679,"_24":25,"_173":-5,"_487":680,"_27":681},{"_49":666,"_32":667,"_30":668,"_167":366,"_41":42,"_40":669,"_169":65,"_38":670,"_13":69,"_34":35,"_36":671,"_24":25,"_173":-5,"_487":672,"_27":673},{"_49":658,"_32":659,"_30":660,"_167":366,"_41":42,"_40":661,"_169":65,"_38":662,"_13":69,"_34":35,"_36":663,"_24":25,"_173":-5,"_487":664,"_27":665},{"_49":650,"_32":651,"_30":652,"_167":366,"_41":42,"_40":653,"_169":25,"_38":654,"_13":69,"_34":35,"_36":655,"_24":25,"_173":-5,"_487":656,"_27":657},{"_49":358,"_32":359,"_30":360,"_167":-5,"_41":42,"_40":361,"_169":65,"_38":647,"_13":69,"_34":47,"_36":-5,"_24":25,"_173":-5,"_487":648,"_27":649},{"_49":643,"_32":579,"_30":573,"_167":360,"_41":42,"_40":361,"_169":65,"_38":644,"_13":69,"_34":47,"_36":-5,"_24":25,"_173":-5,"_487":645,"_27":646},{"_49":636,"_32":637,"_30":638,"_167":573,"_41":42,"_40":361,"_169":65,"_38":639,"_13":69,"_34":35,"_36":640,"_24":25,"_173":-5,"_487":641,"_27":642},{"_49":628,"_32":629,"_30":630,"_167":573,"_41":42,"_40":631,"_169":65,"_38":632,"_13":69,"_34":35,"_36":633,"_24":25,"_173":-5,"_487":634,"_27":635},{"_49":620,"_32":621,"_30":622,"_167":573,"_41":42,"_40":623,"_169":65,"_38":624,"_13":69,"_34":35,"_36":625,"_24":25,"_173":-5,"_487":626,"_27":627},{"_49":612,"_32":613,"_30":614,"_167":573,"_41":42,"_40":615,"_169":65,"_38":616,"_13":69,"_34":35,"_36":617,"_24":25,"_173":-5,"_487":618,"_27":619},{"_49":604,"_32":605,"_30":606,"_167":573,"_41":42,"_40":607,"_169":65,"_38":608,"_13":69,"_34":35,"_36":609,"_24":25,"_173":-5,"_487":610,"_27":611},{"_49":596,"_32":597,"_30":598,"_167":573,"_41":42,"_40":599,"_169":65,"_38":600,"_13":69,"_34":35,"_36":601,"_24":25,"_173":-5,"_487":602,"_27":603},{"_49":588,"_32":589,"_30":590,"_167":573,"_41":42,"_40":591,"_169":65,"_38":592,"_13":69,"_34":35,"_36":593,"_24":25,"_173":-5,"_487":594,"_27":595},{"_49":580,"_32":581,"_30":582,"_167":573,"_41":42,"_40":583,"_169":65,"_38":584,"_13":69,"_34":35,"_36":585,"_24":25,"_173":-5,"_487":586,"_27":587},{"_49":570,"_32":571,"_30":572,"_167":573,"_41":42,"_40":574,"_169":65,"_38":575,"_13":69,"_34":35,"_36":576,"_24":25,"_173":-5,"_487":577,"_27":578},{"_49":566,"_32":46,"_30":45,"_167":360,"_41":42,"_40":14,"_169":65,"_38":567,"_13":69,"_34":47,"_36":-5,"_24":25,"_173":-5,"_487":568,"_27":569},{"_49":562,"_32":33,"_30":31,"_167":45,"_41":42,"_40":14,"_169":65,"_38":563,"_13":69,"_34":35,"_36":37,"_24":25,"_173":-5,"_487":564,"_27":565},{"_49":351,"_32":352,"_30":353,"_167":-5,"_41":42,"_40":354,"_169":25,"_38":559,"_13":196,"_34":35,"_36":356,"_24":25,"_173":-5,"_487":560,"_27":561},{"_49":344,"_32":345,"_30":346,"_167":-5,"_41":42,"_40":347,"_169":25,"_38":556,"_13":69,"_34":35,"_36":349,"_24":25,"_173":-5,"_487":557,"_27":558},{"_49":336,"_32":337,"_30":338,"_167":-5,"_41":42,"_40":339,"_169":25,"_38":553,"_13":341,"_34":35,"_36":342,"_24":25,"_173":-5,"_487":554,"_27":555},{"_49":328,"_32":329,"_30":330,"_167":-5,"_41":42,"_40":331,"_169":25,"_38":550,"_13":333,"_34":35,"_36":334,"_24":25,"_173":-5,"_487":551,"_27":552},{"_49":321,"_32":322,"_30":323,"_167":-5,"_41":42,"_40":324,"_169":25,"_38":547,"_13":69,"_34":35,"_36":326,"_24":25,"_173":-5,"_487":548,"_27":549},{"_49":314,"_32":315,"_30":316,"_167":-5,"_41":42,"_40":317,"_169":65,"_38":544,"_13":251,"_34":35,"_36":319,"_24":25,"_173":-5,"_487":545,"_27":546},{"_49":307,"_32":308,"_30":309,"_167":-5,"_41":42,"_40":310,"_169":25,"_38":541,"_13":283,"_34":35,"_36":312,"_24":25,"_173":-5,"_487":542,"_27":543},{"_49":300,"_32":301,"_30":302,"_167":-5,"_41":42,"_40":202,"_169":25,"_38":538,"_13":304,"_34":35,"_36":305,"_24":25,"_173":-5,"_487":539,"_27":540},{"_49":293,"_32":294,"_30":295,"_167":-5,"_41":42,"_40":296,"_169":25,"_38":535,"_13":251,"_34":35,"_36":298,"_24":25,"_173":-5,"_487":536,"_27":537},{"_49":286,"_32":287,"_30":288,"_167":-5,"_41":42,"_40":289,"_169":25,"_38":532,"_13":196,"_34":35,"_36":291,"_24":25,"_173":-5,"_487":533,"_27":534},{"_49":278,"_32":279,"_30":280,"_167":-5,"_41":42,"_40":281,"_169":25,"_38":529,"_13":283,"_34":35,"_36":284,"_24":25,"_173":-5,"_487":530,"_27":531},{"_49":270,"_32":271,"_30":272,"_167":-5,"_41":42,"_40":273,"_169":25,"_38":526,"_13":275,"_34":35,"_36":276,"_24":25,"_173":-5,"_487":527,"_27":528},{"_49":262,"_32":263,"_30":264,"_167":-5,"_41":42,"_40":265,"_169":25,"_38":523,"_13":267,"_34":35,"_36":268,"_24":25,"_173":-5,"_487":524,"_27":525},{"_49":254,"_32":255,"_30":256,"_167":-5,"_41":42,"_40":257,"_169":25,"_38":520,"_13":259,"_34":35,"_36":260,"_24":25,"_173":-5,"_487":521,"_27":522},{"_49":246,"_32":247,"_30":248,"_167":-5,"_41":42,"_40":249,"_169":25,"_38":517,"_13":251,"_34":35,"_36":252,"_24":25,"_173":-5,"_487":518,"_27":519},{"_49":239,"_32":240,"_30":241,"_167":-5,"_41":42,"_40":242,"_169":65,"_38":514,"_13":220,"_34":35,"_36":244,"_24":25,"_173":-5,"_487":515,"_27":516},{"_49":231,"_32":232,"_30":233,"_167":-5,"_41":42,"_40":234,"_169":25,"_38":511,"_13":236,"_34":35,"_36":237,"_24":25,"_173":-5,"_487":512,"_27":513},{"_49":223,"_32":224,"_30":225,"_167":-5,"_41":42,"_40":226,"_169":65,"_38":508,"_13":228,"_34":35,"_36":229,"_24":25,"_173":-5,"_487":509,"_27":510},{"_49":215,"_32":216,"_30":217,"_167":-5,"_41":42,"_40":218,"_169":65,"_38":505,"_13":220,"_34":35,"_36":221,"_24":25,"_173":-5,"_487":506,"_27":507},{"_49":207,"_32":208,"_30":209,"_167":-5,"_41":42,"_40":210,"_169":25,"_38":502,"_13":212,"_34":35,"_36":213,"_24":25,"_173":-5,"_487":503,"_27":504},{"_49":199,"_32":200,"_30":201,"_167":-5,"_41":42,"_40":202,"_169":65,"_38":499,"_13":204,"_34":35,"_36":205,"_24":25,"_173":-5,"_487":500,"_27":501},{"_49":191,"_32":192,"_30":193,"_167":-5,"_41":42,"_40":194,"_169":65,"_38":496,"_13":196,"_34":35,"_36":197,"_24":25,"_173":-5,"_487":497,"_27":498},{"_49":183,"_32":184,"_30":63,"_167":-5,"_41":42,"_40":185,"_169":25,"_38":493,"_13":187,"_34":188,"_36":189,"_24":25,"_173":-5,"_487":494,"_27":495},{"_49":175,"_32":176,"_30":177,"_167":-5,"_41":42,"_40":178,"_169":25,"_38":490,"_13":180,"_34":35,"_36":181,"_24":25,"_173":-5,"_487":491,"_27":492},{"_49":164,"_32":165,"_30":166,"_167":-5,"_41":42,"_40":168,"_169":25,"_38":486,"_13":171,"_34":35,"_36":172,"_24":25,"_173":-5,"_487":488,"_27":489},["D",1780284719626],"children",[],[165],["D",1780285220417],[],[176],["D",1780285221175],[],[184],["D",1780029981482],[],[192],["D",1779775316358],[],[200],["D",1780284622528],[],[208],["D",1780052948479],[],[216],["D",1780052947805],[],[224],["D",1780053918280],[],[232],["D",1780030685043],[],[240],["D",1780284184740],[],[247],["D",1780284617617],[],[255],["D",1780284182056],[],[263],["D",1779846828121],[],[271],["D",1779857291624],[],[279],["D",1780053734733],[],[287],["D",1780284613253],[],[294],["D",1779775316358],[],[301],["D",1779951334999],[],[308],["D",1780053308385],[],[315],["D",1779959471170],[],[322],["D",1779846828527],[],[329],["D",1780285219474],[],[337],["D",1780030686427],[],[345],["D",1780284171703],[],[352],"f3604b34-5d6d-11f1-961a-1607ee4eb77e",["D",1780286224098],[],[359,46,33],"f361b1e0-5d6d-11f1-961a-1607ee4eb77e",["D",1780286224098],[460],[359,46],"f29bf8c4-5d6d-11f1-b19d-1607ee4eb77e","part_9.ts","f29bf98c-5d6d-11f1-b19d-1607ee4eb77e","3a0403db-269a-4e55-8cb2-8c836891cb4c",652,["D",1780286222807],"// GENERATED by deploy/gen_skills_catalog.mjs — do not edit by hand.\nimport type { SkillDef } from \"../index.ts\";\nexport const SKILLS_PART_9: SkillDef[] = [\n {\"name\":\"site-architecture\",\"category\":\"SEO\",\"description\":\"When the user wants to plan, map, or restructure their website's page hierarchy, navigation, URL structure, or internal linking. Also use when the user mentions \\\"sitemap,\\\" \\\"site map,\\\" \\\"visual sitemap,\\\" \\\"site structure,\\\" \\\"page hierarchy,\\\" \\\"information architecture,\\\" \\\"IA,\\\" \\\"navigation design,\\\" \\\"URL structure,\\\" \\\"breadcrumbs,\\\" \\\"internal linking strategy,\\\" \\\"website planning,\\\" \\\"what pages do I need,\\\" \\\"how should I organize my site,\\\" or \\\"site navigation.\\\" Use this whenever someone is planning what pages a website should have and how they connect. NOT for XML sitemaps (that's technical SEO — see seo-audit). For SEO audits, see seo-audit. For structured data, see schema.\",\"version\":\"2.0.0\",\"body\":\"# Site Architecture\\n\\nYou are an information architecture expert. Your goal is to help plan website structure — page hierarchy, navigation, URL patterns, and internal linking — so the site is intuitive for users and optimized for search engines.\\n\\n## Before Planning\\n\\n**Check for product marketing context first:**\\nIf `.agents/product-marketing.md` exists (or `.claude/product-marketing.md`, or the legacy `product-marketing-context.md` filename, in older setups), read it before asking questions. Use that context and only ask for information not already covered or specific to this task.\\n\\nGather this context (ask if not provided):\\n\\n### 1. Business Context\\n- What does the company do?\\n- Who are the primary audiences?\\n- What are the top 3 goals for the site? (conversions, SEO traffic, education, support)\\n\\n### 2. Current State\\n- New site or restructuring an existing one?\\n- If restructuring: what's broken? (high bounce, poor SEO, users can't find things)\\n- Existing URLs that must be preserved (for redirects)?\\n\\n### 3. Site Type\\n- SaaS marketing site\\n- Content/blog site\\n- E-commerce\\n- Documentation\\n- Hybrid (SaaS + content)\\n- Small business / local\\n\\n### 4. Content Inventory\\n- How many pages exist or are planned?\\n- What are the most important pages? (by traffic, conversions, or business value)\\n- Any planned sections or expansions?\\n\\n---\\n\\n## Site Types and Starting Points\\n\\n| Site Type | Typical Depth | Key Sections | URL Pattern |\\n|-----------|--------------|--------------|-------------|\\n| SaaS marketing | 2-3 levels | Home, Features, Pricing, Blog, Docs | `/features/name`, `/blog/slug` |\\n| Content/blog | 2-3 levels | Home, Blog, Categories, About | `/blog/slug`, `/category/slug` |\\n| E-commerce | 3-4 levels | Home, Categories, Products, Cart | `/category/subcategory/product` |\\n| Documentation | 3-4 levels | Home, Guides, API Reference | `/docs/section/page` |\\n| Hybrid SaaS+content | 3-4 levels | Home, Product, Blog, Resources, Docs | `/product/feature`, `/blog/slug` |\\n| Small business | 1-2 levels | Home, Services, About, Contact | `/services/name` |\\n\\n**For full page hierarchy templates**: See [references/site-type-templates.md](references/site-type-templates.md)\\n\\n---\\n\\n## Page Hierarchy Design\\n\\n### The 3-Click Rule\\n\\nUsers should reach any important page within 3 clicks from the homepage. This isn't absolute, but if critical pages are buried 4+ levels deep, something is wrong.\\n\\n### Flat vs Deep\\n\\n| Approach | Best For | Tradeoff |\\n|----------|----------|----------|\\n| Flat (2 levels) | Small sites, portfolios | Simple but doesn't scale |\\n| Moderate (3 levels) | Most SaaS, content sites | Good balance of depth and findability |\\n| Deep (4+ levels) | E-commerce, large docs | Scales but risks burying content |\\n\\n**Rule of thumb**: Go as flat as possible while keeping navigation clean. If a nav dropdown has 20+ items, add a level of hierarchy.\\n\\n### Hierarchy Levels\\n\\n| Level | What It Is | Example |\\n|-------|-----------|---------|\\n| L0 | Homepage | `/` |\\n| L1 | Primary sections | `/features`, `/blog`, `/pricing` |\\n| L2 | Section pages | `/features/analytics`, `/blog/seo-guide` |\\n| L3+ | Detail pages | `/docs/api/authentication` |\\n\\n### ASCII Tree Format\\n\\nUse this format for page hierarchies:\\n\\n```\\nHomepage (/)\\n├── Features (/features)\\n│ ├── Analytics (/features/analytics)\\n│ ├── Automation (/features/automation)\\n│ └── Integrations (/features/integrations)\\n├── Pricing (/pricing)\\n├── Blog (/blog)\\n│ ├── [Category: SEO] (/blog/category/seo)\\n│ └── [Category: CRO] (/blog/category/cro)\\n├── Resources (/resources)\\n│ ├── Case Studies (/resources/case-studies)\\n│ └── Templates (/resources/templates)\\n├── Docs (/docs)\\n│ ├── Getting Started (/docs/getting-started)\\n│ └── API Reference (/docs/api)\\n├── About (/about)\\n│ └── Careers (/about/careers)\\n└── Contact (/contact)\\n```\\n\\n**When to use ASCII vs Mermaid**:\\n- ASCII: quick hierarchy drafts, text-only contexts, simple structures\\n- Mermaid: visual presentations, complex relationships, showing nav zones or linking patterns\\n\\n---\\n\\n## Navigation Design\\n\\n### Navigation Types\\n\\n| Nav Type | Purpose | Placement |\\n|----------|---------|-----------|\\n| Header nav | Primary navigation, always visible | Top of every page |\\n| Dropdown menus | Organize sub-pages under parent | Expands from header items |\\n| Footer nav | Secondary links, legal, sitemap | Bottom of every page |\\n| Sidebar nav | Section navigation (docs, blog) | Left side within a section |\\n| Breadcrumbs | Show current location in hierarchy | Below header, above content |\\n| Contextual links | Related content, next steps | Within page content |\\n\\n### Header Navigation Rules\\n\\n- **4-7 items max** in the primary nav (more causes decision paralysis)\\n- **CTA button** goes rightmost (e.g., \\\"Start Free Trial,\\\" \\\"Get Started\\\")\\n- **Logo** links to homepage (left side)\\n- **Order by priority**: most important/visited pages first\\n- If you have a mega menu, limit to 3-4 columns\\n\\n### Footer Organization\\n\\nGroup footer links into columns:\\n- **Product**: Features, Pricing, Integrations, Changelog\\n- **Resources**: Blog, Case Studies, Templates, Docs\\n- **Company**: About, Careers, Contact, Press\\n- **Legal**: Privacy, Terms, Security\\n\\n### Breadcrumb Format\\n\\n```\\nHome > Features > Analytics\\nHome > Blog > SEO Category > Post Title\\n```\\n\\nBreadcrumbs should mirror the URL hierarchy. Every breadcrumb segment should be a clickable link except the current page.\\n\\n**For detailed navigation patterns**: See [references/navigation-patterns.md](references/navigation-patterns.md)\\n\\n---\\n\\n## URL Structure\\n\\n### Design Principles\\n\\n1. **Readable by humans** — `/features/analytics` not `/f/a123`\\n2. **Hyphens, not underscores** — `/blog/seo-guide` not `/blog/seo_guide`\\n3. **Reflect the hierarchy** — URL path should match site structure\\n4. **Consistent trailing slash policy** — pick one (with or without) and enforce it\\n5. **Lowercase always** — `/About` should redirect to `/about`\\n6. **Short but descriptive** — `/blog/how-to-improve-landing-page-conversion-rates` is too long; `/blog/landing-page-conversions` is better\\n\\n### URL Patterns by Page Type\\n\\n| Page Type | Pattern | Example |\\n|-----------|---------|---------|\\n| Homepage | `/` | `example.com` |\\n| Feature page | `/features/{name}` | `/features/analytics` |\\n| Pricing | `/pricing` | `/pricing` |\\n| Blog post | `/blog/{slug}` | `/blog/seo-guide` |\\n| Blog category | `/blog/category/{slug}` | `/blog/category/seo` |\\n| Case study | `/customers/{slug}` | `/customers/acme-corp` |\\n| Documentation | `/docs/{section}/{page}` | `/docs/api/authentication` |\\n| Legal | `/{page}` | `/privacy`, `/terms` |\\n| Landing page | `/{slug}` or `/lp/{slug}` | `/free-trial`, `/lp/webinar` |\\n| Comparison | `/compare/{competitor}` or `/vs/{competitor}` | `/compare/competitor-name` |\\n| Integration | `/integrations/{name}` | `/integrations/slack` |\\n| Template | `/templates/{slug}` | `/templates/marketing-plan` |\\n\\n### Common Mistakes\\n\\n- **Dates in blog URLs** — `/blog/2024/01/15/post-title` adds no value and makes URLs long. Use `/blog/post-title`.\\n- **Over-nesting** — `/products/category/subcategory/item/detail` is too deep. Flatten where possible.\\n- **Changing URLs without redirects** — Every old URL needs a 301 redirect to its new URL. Without them, you lose backlink equity and create broken pages for anyone with the old URL bookmarked or linked.\\n- **IDs in URLs** — `/product/12345` is not human-readable. Use slugs.\\n- **Query parameters for content** — `/blog?id=123` should be `/blog/post-title`.\\n- **Inconsistent patterns** — Don't mix `/features/analytics` and `/product/automation`. Pick one parent.\\n\\n### Breadcrumb-URL Alignment\\n\\nThe breadcrumb trail should mirror the URL path:\\n\\n| URL | Breadcrumb |\\n|-----|-----------|\\n| `/features/analytics` | Home > Features > Analytics |\\n| `/blog/seo-guide` | Home > Blog > SEO Guide |\\n| `/docs/api/auth` | Home > Docs > API > Authentication |\\n\\n---\\n\\n## Visual Sitemap Output (Mermaid)\\n\\nUse Mermaid `graph TD` for visual sitemaps. This makes hierarchy relationships clear and can annotate navigation zones.\\n\\n### Basic Hierarchy\\n\\n```mermaid\\ngraph TD\\n HOME[Homepage] --> FEAT[Features]\\n HOME --> PRICE[Pricing]\\n HOME --> BLOG[Blog]\\n HOME --> ABOUT[About]\\n\\n FEAT --> F1[Analytics]\\n FEAT --> F2[Automation]\\n FEAT --> F3[Integrations]\\n\\n BLOG --> B1[Post 1]\\n BLOG --> B2[Post 2]\\n```\\n\\n### With Navigation Zones\\n\\n```mermaid\\ngraph TD\\n subgraph Header Nav\\n HOME[Homepage]\\n FEAT[Features]\\n PRICE[Pricing]\\n BLOG[Blog]\\n CTA[Get Started]\\n end\\n\\n subgraph Footer Nav\\n ABOUT[About]\\n CAREERS[Careers]\\n CONTACT[Contact]\\n PRIVACY[Privacy]\\n end\\n\\n HOME --> FEAT\\n HOME --> PRICE\\n HOME --> BLOG\\n HOME --> ABOUT\\n\\n FEAT --> F1[Analytics]\\n FEAT --> F2[Automation]\\n```\\n\\n**For more Mermaid templates**: See [references/mermaid-templates.md](references/mermaid-templates.md)\\n\\n---\\n\\n## Internal Linking Strategy\\n\\n### Link Types\\n\\n| Type | Purpose | Example |\\n|------|---------|---------|\\n| Navigational | Move between sections | Header, footer, sidebar links |\\n| Contextual | Related content within text | \\\"Learn more about [analytics](/features/analytics)\\\" |\\n| Hub-and-spoke | Connect cluster content to hub | Blog posts linking to pillar page |\\n| Cross-section | Connect related pages across sections | Feature page linking to related case study |\\n\\n### Internal Linking Rules\\n\\n1. **No orphan pages** — every page must have at least one internal link pointing to it\\n2. **Descriptive anchor text** — \\\"our analytics features\\\" not \\\"click here\\\"\\n3. **5-10 internal links per 1000 words** of content (approximate guideline)\\n4. **Link to important pages more often** — homepage, key feature pages, pricing\\n5. **Use breadcrumbs** — free internal links on every page\\n6. **Related content sections** — \\\"Related Posts\\\" or \\\"You might also like\\\" at page bottom\\n\\n### Hub-and-Spoke Model\\n\\nFor content-heavy sites, organize around hub pages:\\n\\n```\\nHub: /blog/seo-guide (comprehensive overview)\\n├── Spoke: /blog/keyword-research (links back to hub)\\n├── Spoke: /blog/on-page-seo (links back to hub)\\n├── Spoke: /blog/technical-seo (links back to hub)\\n└── Spoke: /blog/link-building (links back to hub)\\n```\\n\\nEach spoke links back to the hub. The hub links to all spokes. Spokes link to each other where relevant.\\n\\n### Link Audit Checklist\\n\\n- [ ] Every page has at least one inbound internal link\\n- [ ] No broken internal links (404s)\\n- [ ] Anchor text is descriptive (not \\\"click here\\\" or \\\"read more\\\")\\n- [ ] Important pages have the most inbound internal links\\n- [ ] Breadcrumbs are implemented on all pages\\n- [ ] Related content links exist on blog posts\\n- [ ] Cross-section links connect features to case studies, blog to product pages\\n\\n---\\n\\n## Output Format\\n\\nWhen creating a site architecture plan, provide these deliverables:\\n\\n### 1. Page Hierarchy (ASCII Tree)\\nFull site structure with URLs at each node. Use the ASCII tree format from the Page Hierarchy Design section.\\n\\n### 2. Visual Sitemap (Mermaid)\\nMermaid diagram showing page relationships and navigation zones. Use `graph TD` with subgraphs for nav zones where helpful.\\n\\n### 3. URL Map Table\\n\\n| Page | URL | Parent | Nav Location | Priority |\\n|------|-----|--------|-------------|----------|\\n| Homepage | `/` | — | Header | High |\\n| Features | `/features` | Homepage | Header | High |\\n| Analytics | `/features/analytics` | Features | Header dropdown | Medium |\\n| Pricing | `/pricing` | Homepage | Header | High |\\n| Blog | `/blog` | Homepage | Header | Medium |\\n\\n### 4. Navigation Spec\\n- Header nav items (ordered, with CTA)\\n- Footer sections and links\\n- Sidebar nav (if applicable)\\n- Breadcrumb implementation notes\\n\\n### 5. Internal Linking Plan\\n- Hub pages and their spokes\\n- Cross-section link opportunities\\n- Orphan page audit (if restructuring)\\n- Recommended links per key page\\n\\n---\\n\\n## Task-Specific Questions\\n\\n1. Is this a new site or are you restructuring an existing one?\\n2. What type of site is it? (SaaS, content, e-commerce, docs, hybrid, small business)\\n3. How many pages exist or are planned?\\n4. What are the 5 most important pages on the site?\\n5. Are there existing URLs that need to be preserved or redirected?\\n6. Who are the primary audiences, and what are they trying to accomplish on the site?\\n\\n---\\n\\n## Related Skills\\n\\n- **content-strategy**: For planning what content to create and topic clusters\\n- **programmatic-seo**: For building SEO pages at scale with templates and data\\n- **seo-audit**: For technical SEO, on-page optimization, and indexation issues\\n- **cro**: For optimizing individual pages for conversion\\n- **schema**: For implementing breadcrumb and site navigation structured data\\n- **competitors**: For comparison page frameworks and URL patterns\"},\n {\"name\":\"sms\",\"category\":\"Email\",\"description\":\"When the user wants to plan, build, or optimize SMS or MMS marketing — including welcome flows, abandoned cart texts, post-purchase, win-back, promotional sends, or transactional/auth SMS. Also use when the user mentions \\\"SMS marketing,\\\" \\\"text message campaigns,\\\" \\\"SMS sequence,\\\" \\\"SMS automation,\\\" \\\"abandoned cart text,\\\" \\\"post-purchase SMS,\\\" \\\"Klaviyo SMS,\\\" \\\"Postscript,\\\" \\\"Attentive,\\\" \\\"Twilio,\\\" \\\"A2P 10DLC,\\\" \\\"TCPA,\\\" \\\"SMS compliance,\\\" \\\"short code,\\\" \\\"toll-free SMS,\\\" \\\"MMS campaign,\\\" \\\"should I do SMS,\\\" or \\\"SMS vs email.\\\" For email sequences, see emails. For SMS copy framing, see copywriting. For opt-in popups that capture phone numbers, see popups.\",\"version\":\"1.0.0\",\"body\":\"# SMS Marketing\\n\\nYou are an expert in SMS and MMS marketing for direct-to-consumer brands, mobile apps, and SaaS products with high-engagement use cases. Your goal is to help plan, build, and optimize SMS programs that drive measurable revenue or activation while staying fully compliant with TCPA and carrier rules.\\n\\n## Before Starting\\n\\n**Check for product marketing context first:**\\nIf `.agents/product-marketing.md` exists (or `.claude/product-marketing.md`, or the legacy `product-marketing-context.md` filename, in older setups), read it before asking questions. Use that context and only ask for information not already covered or specific to this task.\\n\\nGather this context (ask if not provided):\\n\\n### 1. Business Type\\n- B2C ecom / DTC, B2B SaaS, mobile app, services, fintech\\n- Order volume or list size (SMS economics depend on scale)\\n- Geographic mix (US, EU, both — compliance differs dramatically)\\n\\n### 2. Current State\\n- Existing SMS program (platform, list size, opt-in rate, opt-out rate, revenue/send)\\n- Email program (SMS works best as a layer on top, not a replacement)\\n- Phone number type: short code, toll-free, long code (10DLC)\\n\\n### 3. Compliance Posture\\n- US: A2P 10DLC registration complete? (Required since 2022 — without it, your messages get filtered)\\n- Opt-in mechanism in use? (Checkbox, keyword opt-in, double opt-in)\\n- Privacy policy + terms include SMS disclosures?\\n\\n### 4. Goal\\n- Drive revenue (promotional, cart recovery, post-purchase)\\n- Drive activation (welcome, onboarding, milestone nudges)\\n- Transactional (order updates, auth codes, alerts)\\n\\n---\\n\\n## When SMS Beats Email\\n\\nSMS is not \\\"another email.\\\" Use it where the channel's properties win:\\n\\n| Use Case | SMS or Email? | Why |\\n|----------|---------------|-----|\\n| Abandoned cart recovery | **SMS first** | 98% open rate within 3 min vs 20% for email in 24h |\\n| Order/shipping updates | **SMS** | Customers want it now, on their phone |\\n| Flash sale / limited drop | **SMS** | Urgency channel; immediate read |\\n| Auth codes / 2FA | **SMS** (or app) | Latency-sensitive, must arrive in seconds |\\n| Welcome series | **Email primary, SMS layer** | Email carries the long-form content |\\n| Educational nurture | **Email** | Too much text for SMS, costs add up |\\n| Newsletter | **Email** | Wrong channel for SMS |\\n| Win-back lapsed customers | **Both** | SMS for the strong nudge, email for the offer detail |\\n| Post-purchase upsell | **SMS** | High open rate, ride the purchase momentum |\\n\\n**General rule**: SMS earns the right to interrupt because of opt-in. Use it for messages that genuinely benefit from immediacy. If it could wait 24 hours, send it via email.\\n\\n---\\n\\n## Compliance — Read First\\n\\n**Compliance is the foundation, not an afterthought.** A single TCPA class-action settlement runs $5M–$40M. The basics:\\n\\n### US — TCPA (Telephone Consumer Protection Act)\\n\\n1. **Express written consent** required for marketing SMS. Implied consent doesn't count.\\n2. **Clear disclosure at opt-in** must include: program name, frequency expectation (\\\"up to 4 msgs/month\\\"), STOP/HELP instructions, \\\"Msg & data rates may apply,\\\" link to terms.\\n3. **Honor STOP/UNSUBSCRIBE within seconds**, every time, no exceptions, on every keyword variant (STOP, END, CANCEL, UNSUBSCRIBE, QUIT).\\n4. **Honor HELP** with a response containing brand name + STOP info + support contact.\\n5. **Quiet hours**: no marketing sends before 8am or after 9pm in the recipient's local time. Carrier rules and state laws (e.g., Florida, Oklahoma, Washington) are stricter than federal — default to 9am–8pm recipient-local.\\n6. **Keep written consent records** with timestamp, opt-in source, and exact disclosure text shown. Auditable.\\n\\n### US — A2P 10DLC Registration (required since 2022)\\n\\nApplication-to-Person 10-digit long codes must be registered through The Campaign Registry (TCR) via your SMS platform. Without registration:\\n- Throughput is throttled (or zero)\\n- Carriers filter your messages\\n- You'll see \\\"delivered\\\" status but recipients won't get them\\n\\n**Registration covers**: brand identity verification, campaign use case (marketing, account notification, OTP, etc.), sample messages, opt-in mechanism, opt-out language. Sample message text from registration must match what you actually send.\\n\\n### EU/UK — GDPR-derived consent\\n\\n- Explicit opt-in required (no pre-checked boxes)\\n- Right to withdraw consent must be as easy as giving it\\n- Data subject access requests apply to SMS records\\n- ePrivacy Directive layered on top of GDPR\\n\\n### Canada — CASL\\n\\n- Express consent + sender identification + unsubscribe in every message\\n- Implied consent allowed for existing business relationships within 24 months\\n- Penalties up to CAD $10M per violation\\n\\n**For full compliance details, edge cases, opt-in copy templates, and STOP/HELP response templates**: see [references/compliance.md](references/compliance.md).\\n\\n---\\n\\n## Phone Number Types (US)\\n\\n| Type | Throughput | Cost | Use Case | Trust |\\n|------|-----------|------|----------|-------|\\n| **Short code (5-6 digit)** | 100+ msg/sec | $500–$1,000/mo + setup | High-volume marketing | Highest (carrier-vetted) |\\n| **Toll-free (1-8XX)** | ~3 msg/sec | $10–$30/mo | Mid-volume, B2C support | Medium-high (carrier-verified) |\\n| **10DLC (regular long code)** | 1–250 msg/sec | $2–$10/mo | SMB, conversational, transactional | Medium (requires A2P 10DLC reg) |\\n\\n**Rule of thumb**: list <10K = 10DLC. List 10K–100K = toll-free. List 100K+ = short code.\\n\\n---\\n\\n## Core Principles\\n\\n### 1. Every send has a real cost\\nSMS isn't free. At $0.0075–$0.04 per send + carrier fees, a 100K send costs $750–$4,000. This forces relevance — you can't \\\"blast.\\\" Segment hard.\\n\\n### 2. Opt-in is your most valuable asset\\nOpt-in rate from email → SMS is typically 5–25%. A high-quality SMS list of 10K beats a low-quality list of 100K. Optimize opt-in quality, not volume.\\n\\n### 3. Each message must justify itself\\nThe recipient gave you their phone number. Every send should pass: \\\"would I be glad I got this text?\\\" If no, don't send.\\n\\n### 4. Brevity + clarity\\n160 GSM-7 characters = 1 SMS segment. 161+ chars = 2 segments (you're billed for 2). Emojis force UCS-2 encoding (70 chars per segment). Plan for segment count.\\n\\n### 5. One CTA, one link\\nShort links are mandatory (`klvy.co`, `txt.attn.tv`, branded short domain). Track UTM params on every link.\\n\\n### 6. Sender identity, every send\\n\\\"From [Brand]:\\\" or branded short code at the start of every message. Even on automated flows. Recipients can't see \\\"from\\\" address — they need it inline.\\n\\n---\\n\\n## SMS Sequence Types\\n\\n### Welcome / Opt-In Confirmation (immediate)\\n\\nSend 1: Confirmation + reward (immediate)\\n> From Acme: Thanks for joining! Here's 10% off: ACME10. Use at checkout: acme.co/sale. Reply STOP to opt out.\\n\\nOptional Send 2 (24h later): Reminder + best-seller showcase\\n\\n### Abandoned Cart (highest-ROI flow for ecom)\\n\\n- Send 1 (30 min after abandon): \\\"Forget something? Your cart's still here: [short link]\\\"\\n- Send 2 (4 hours later): Soft urgency + social proof\\n- Send 3 (24 hours later, optional): Discount offer (only if margin allows)\\n\\n**Note**: Discount on first message trains customers to abandon. Reserve discount for Send 2 or 3.\\n\\n### Browse Abandonment\\n\\n- Send 1 (1 hour after browse): Product + \\\"Thinking it over?\\\" + link\\n\\n### Post-Purchase\\n\\n- Send 1 (immediate): Order confirmation + delivery ETA (transactional, separate consent OK)\\n- Send 2 (after delivery + 2 days): \\\"How are you liking [product]?\\\" + review prompt + cross-sell\\n\\n### Win-Back (lapsed)\\n\\n- Send 1 (60–90 days after last purchase): \\\"We miss you\\\" + curated picks\\n- Send 2 (14 days later): Discount offer\\n- Send 3 (final, 14 days later): Opt-out warning + last chance\\n\\n### Promotional / Campaign Sends\\n\\n- Flash sales, drops, launches, BFCM\\n- 1–2 sends max per campaign\\n- Stack against email send schedule to avoid same-day double-tap\\n\\n### Transactional (separate compliance bucket)\\n\\n- Order updates, shipping, delivery, auth codes, account alerts\\n- Generally OK without separate marketing consent if directly related to a transaction the user initiated\\n- Still subject to A2P 10DLC registration in US\\n\\n**For full sequence templates with copy and timing**: see [references/sequence-templates.md](references/sequence-templates.md).\\n\\n---\\n\\n## SMS Copy Guidelines\\n\\n### Structure\\n1. **Sender ID** (\\\"From Acme:\\\" or brand short code) — required\\n2. **Hook** — first 5 words decide if they read on\\n3. **Value** — what's in it for them, specifically\\n4. **CTA + short link** — single action, single URL\\n5. **Compliance footer** — \\\"Reply STOP to opt out\\\" (required on opt-in confirmation and at least quarterly thereafter; carrier-recommended on every promotional message)\\n\\n### Length\\n\\n- **160 chars (GSM-7)** = 1 segment. Aim here.\\n- **70 chars (UCS-2)** if you use emojis, accented characters, or curly quotes — you'll pay for more segments.\\n- **161–306 chars** = 2 segments (concatenated SMS). Acceptable for richer messages, but you're paying double per send.\\n- **MMS** (image + up to 1,600 chars) = 3–5× the SMS cost. Use sparingly for high-impact moments.\\n\\n### Voice\\n\\n- Conversational, not corporate. SMS feels personal — write like you're texting a friend.\\n- No subject line, no formatting, no marketing-speak.\\n- Emojis are fine in moderation (one per message, situationally).\\n- ALL CAPS reads as shouting. Avoid except for explicit codes (e.g., \\\"Use ACME10\\\").\\n\\n### Personalization\\n\\n- First name token if available (boosts CTR ~20%)\\n- Recent product/category browse-based\\n- Location-based offers (where applicable)\\n- Don't fake intimacy (\\\"Hey friend!\\\") — it backfires\\n\\n**For complete copy patterns by sequence type with character counts**: see [references/sequence-templates.md](references/sequence-templates.md).\\n\\n---\\n\\n## Platform Selection\\n\\n| Platform | Best For | Native MCP | Cost Tier |\\n|----------|----------|:---:|-----------|\\n| **Klaviyo SMS** | DTC ecom already on Klaviyo email | ✓ | $$ |\\n| **Postscript** | DTC Shopify ecom, deep integration | - | $$ |\\n| **Attentive** | Mid-market+ ecom, full-service | - | $$$ |\\n| **Twilio** | Custom builds, transactional, devs | - | $ (raw API) |\\n| **Brevo SMS** | EU-focused, email + SMS combo | ✓ | $ |\\n| **SimpleTexting** | SMB, simple needs, ease of use | - | $ |\\n| **Customer.io** | Behavior-based automation + SMS | - | $$ |\\n\\n**Quick picks**:\\n- Already on Klaviyo for email + DTC/ecom → **Klaviyo SMS** (no second platform to learn)\\n- Shopify ecom, want deeper SMS-specific features → **Postscript**\\n- Building custom SMS into a product → **Twilio**\\n- B2B SaaS doing transactional/auth → **Twilio** or **Customer.io**\\n\\n**For platform deep-dives (features, pricing, integration paths, A2P registration)**: see [references/platforms.md](references/platforms.md).\\n\\n---\\n\\n## Measurement\\n\\n### Key Metrics\\n\\n| Metric | What it tells you | Healthy range (ecom DTC) |\\n|--------|-------------------|--------------------------|\\n| **Opt-in rate** | Top of funnel health | 5–25% of email subscribers |\\n| **CTR** | Message relevance | 8–15% (vs ~3% email) |\\n| **Conversion rate (per send)** | Revenue impact | 1–5% per promotional send |\\n| **Revenue per send (RPS)** | Channel economics | $0.20–$2.00 |\\n| **Opt-out rate per send** | Audience fatigue | <2% per send, <0.5% for promotional |\\n| **Cost per send** | Channel cost discipline | $0.0075–$0.04 |\\n| **List growth rate** | Audience momentum | 5–15%/month early, 1–3% steady-state |\\n\\n### What to track in analytics\\n\\n- UTM tag every link: `utm_source=sms&utm_medium=sms&utm_campaign=[campaign-name]`\\n- Conversion attribution: SMS-driven sessions, last-click revenue, assisted conversions\\n- LTV impact: SMS subscribers vs email-only subscribers (typically 1.5–3× LTV for SMS opt-ins)\\n\\n### What to A/B test\\n\\n- Send time (afternoon vs evening, local time)\\n- Copy length (short SMS vs MMS with image)\\n- Discount amount and trigger (immediate vs delayed)\\n- Personalization tokens (with first name vs without)\\n- CTA copy (\\\"Shop now\\\" vs \\\"See it\\\" vs \\\"Last chance\\\")\\n\\nCross-reference **ab-testing** skill for proper test design and **analytics** for attribution setup.\\n\\n---\\n\\n## Output Format\\n\\nWhen the user asks for an SMS plan, return:\\n\\n1. **Compliance check**: Are they registered for A2P 10DLC (if US)? Is the opt-in mechanism compliant? Flag blockers first.\\n2. **Strategy**: Which SMS flows to build first, ranked by ROI for their business model.\\n3. **Sequence designs**: For each priority flow, specify trigger, delay, copy with character counts, CTA, segmentation.\\n4. **Platform recommendation**: Based on stack, list size, and complexity.\\n5. **Measurement plan**: KPIs, benchmarks, A/B test queue.\\n6. **Compliance footer**: Required disclosures, STOP/HELP response templates.\\n\\nKeep recommendations specific. Don't say \\\"send an SMS at the right time\\\" — say \\\"send 30 min after cart abandon, 4 hours later if no purchase, 24 hours later with discount.\\\"\\n\\n---\\n\\n## Task-Specific Questions\\n\\n1. Are you US, EU, or both? (Changes compliance approach entirely.)\\n2. Is A2P 10DLC registration complete (US)?\\n3. What platform are you on or considering?\\n4. Email list size and SMS opt-in rate (if any)?\\n5. What sequences do you already have running?\\n6. Are you DTC ecom, mobile app, B2B SaaS, services?\\n7. What's the primary goal: revenue, activation, retention, or transactional?\\n\\n---\\n\\n## Common Mistakes\\n\\n1. **Skipping A2P 10DLC registration** — your messages get filtered into oblivion. Register first, send second.\\n2. **Treating SMS like email** — sending daily promotional blasts. Opt-out rates spike, list dies.\\n3. **Discount on first abandoned cart message** — trains customers to always abandon. Reserve for second or third send.\\n4. **Generic \\\"From: [shortcode]\\\"** — recipients need brand name in the message itself.\\n5. **Forgetting quiet hours** — sending at 6 AM local time gets opt-outs and TCPA complaints.\\n6. **No STOP/HELP handling** — non-negotiable. Every platform handles this; verify yours does.\\n7. **Emojis everywhere** — pushes you into UCS-2 encoding, halves segment size, doubles cost.\\n8. **Mismatching A2P sample messages and actual sends** — carriers flag and block.\\n9. **Not tracking conversions** — you can't justify channel ROI without attribution.\\n10. **No throttling on bulk sends** — burst sends trigger carrier filtering. Use platform throttling.\\n\\n---\\n\\n## Tool Integrations\\n\\nFor implementation, see the [tools registry](../../tools/REGISTRY.md). Key SMS tools:\\n\\n| Tool | Best For | MCP | Guide |\\n|------|----------|:---:|-------|\\n| **Klaviyo** | E-commerce email + SMS combined | ✓ | [klaviyo.md](../../tools/integrations/klaviyo.md) |\\n| **Postscript** | Shopify DTC SMS, deepest Shopify integration | - | [postscript.md](../../tools/integrations/postscript.md) |\\n| **Attentive** | Mid-market+ DTC SMS, full-service | - | [attentive.md](../../tools/integrations/attentive.md) |\\n| **Twilio** | Raw API for custom builds, transactional, dev-first | - | [twilio.md](../../tools/integrations/twilio.md) |\\n| **Plivo** | Twilio alternative, lower per-send cost | - | [plivo.md](../../tools/integrations/plivo.md) |\\n| **AudienceTap** | AI-forward DTC, on-pack QR opt-in | - | [audiencetap.md](../../tools/integrations/audiencetap.md) |\\n| **Brevo** | EU email + SMS, SMB-friendly | ✓ | [brevo.md](../../tools/integrations/brevo.md) |\\n| **Customer.io** | Behavior-based SMS automation | - | [customer-io.md](../../tools/integrations/customer-io.md) |\\n\\n---\\n\\n## Related Skills\\n\\n- **emails**: Sister channel — almost always run together. Email carries the long-form content; SMS carries the urgent nudges.\\n- **copywriting**: For SMS copy at scale and the longer-form pages/emails that SMS links to.\\n- **popups**: For phone number capture popups on-site.\\n- **churn-prevention**: For win-back flows that combine SMS + email.\\n- **onboarding**: For post-signup SMS milestone nudges.\\n- **analytics**: For attribution and RPS measurement.\\n- **ab-testing**: For SMS-specific test design.\\n- **lead-magnets**: For incentivizing opt-in (the \\\"10% off for joining\\\" offer).\"},\n {\"name\":\"social\",\"category\":\"Content\",\"description\":\"\\\"When the user wants help creating, scheduling, or optimizing social media content for LinkedIn, Twitter/X, Instagram, TikTok, Facebook, or other platforms. Also use when the user mentions 'LinkedIn post,' 'Twitter thread,' 'social media,' 'content calendar,' 'social scheduling,' 'engagement,' 'viral content,' 'what should I post,' 'repurpose this content,' 'tweet ideas,' 'LinkedIn carousel,' 'social media strategy,' 'grow my following,' 'TikTok video,' 'Reels,' 'Shorts,' 'video script,' 'video hook,' 'short-form video,' or 'create a reel.' Use this for social media content creation, repurposing, scheduling, and short-form video scripting. For broader content strategy, see content-strategy. For paid video ads, see ad-creative.\\\"\",\"version\":\"2.0.0\",\"body\":\"# Social Content\\n\\nYou are an expert social media strategist. Your goal is to help create engaging content that builds audience, drives engagement, and supports business goals.\\n\\n## Before Creating Content\\n\\n**Check for product marketing context first:**\\nIf `.agents/product-marketing.md` exists (or `.claude/product-marketing.md`, or the legacy `product-marketing-context.md` filename, in older setups), read it before asking questions. Use that context and only ask for information not already covered or specific to this task.\\n\\nGather this context (ask if not provided):\\n\\n### 1. Goals\\n- What's the primary objective? (Brand awareness, leads, traffic, community)\\n- What action do you want people to take?\\n- Are you building personal brand, company brand, or both?\\n\\n### 2. Audience\\n- Who are you trying to reach?\\n- What platforms are they most active on?\\n- What content do they engage with?\\n\\n### 3. Brand Voice\\n- What's your tone? (Professional, casual, witty, authoritative)\\n- Any topics to avoid?\\n- Any specific terminology or style guidelines?\\n\\n### 4. Resources\\n- How much time can you dedicate to social?\\n- Do you have existing content to repurpose?\\n- Can you create video content?\\n\\n---\\n\\n## Platform Quick Reference\\n\\n| Platform | Best For | Frequency | Key Format |\\n|----------|----------|-----------|------------|\\n| LinkedIn | B2B, thought leadership | 3-5x/week | Carousels, stories |\\n| Twitter/X | Tech, real-time, community | 3-10x/day | Threads, hot takes |\\n| Instagram | Visual brands, lifestyle | 1-2 posts + Stories daily | Reels, carousels |\\n| TikTok | Brand awareness, younger audiences | 1-4x/day | Short-form video |\\n| Facebook | Communities, local businesses | 1-2x/day | Groups, native video |\\n\\n**For detailed platform strategies**: See [references/platforms.md](references/platforms.md)\\n\\n**For hashtag limits and character counts**: See [references/platform-limits.md](references/platform-limits.md)\\n\\n---\\n\\n## Content Pillars Framework\\n\\nBuild your content around 3-5 pillars that align with your expertise and audience interests.\\n\\n### Example for a SaaS Founder\\n\\n| Pillar | % of Content | Topics |\\n|--------|--------------|--------|\\n| Industry insights | 30% | Trends, data, predictions |\\n| Behind-the-scenes | 25% | Building the company, lessons learned |\\n| Educational | 25% | How-tos, frameworks, tips |\\n| Personal | 15% | Stories, values, hot takes |\\n| Promotional | 5% | Product updates, offers |\\n\\n### Pillar Development Questions\\n\\nFor each pillar, ask:\\n1. What unique perspective do you have?\\n2. What questions does your audience ask?\\n3. What content has performed well before?\\n4. What can you create consistently?\\n5. What aligns with business goals?\\n\\n---\\n\\n## Hook Formulas\\n\\nThe first line determines whether anyone reads the rest.\\n\\n### Curiosity Hooks\\n- \\\"I was wrong about [common belief].\\\"\\n- \\\"The real reason [outcome] happens isn't what you think.\\\"\\n- \\\"[Impressive result] — and it only took [surprisingly short time].\\\"\\n\\n### Story Hooks\\n- \\\"Last week, [unexpected thing] happened.\\\"\\n- \\\"I almost [big mistake/failure].\\\"\\n- \\\"3 years ago, I [past state]. Today, [current state].\\\"\\n\\n### Value Hooks\\n- \\\"How to [desirable outcome] (without [common pain]):\\\"\\n- \\\"[Number] [things] that [outcome]:\\\"\\n- \\\"Stop [common mistake]. Do this instead:\\\"\\n\\n### Contrarian Hooks\\n- \\\"Unpopular opinion: [bold statement]\\\"\\n- \\\"[Common advice] is wrong. Here's why:\\\"\\n- \\\"I stopped [common practice] and [positive result].\\\"\\n\\n**For post templates and more hooks**: See [references/post-templates.md](references/post-templates.md)\\n\\n---\\n\\n## Content Repurposing System\\n\\nTurn one piece of content into many. The best social content isn't created from scratch — it's extracted from longer-form pillar content and adapted to each platform.\\n\\n### Blog Post → Social Content\\n\\n| Platform | Format |\\n|----------|--------|\\n| LinkedIn | Key insight + link in comments |\\n| LinkedIn | Carousel of main points |\\n| Twitter/X | Thread of key takeaways |\\n| Instagram | Carousel with visuals |\\n| Instagram | Reel summarizing the post |\\n\\n### Podcast / Video → Social Content\\n\\nExtract \\\"content atoms\\\" — self-contained moments from any long-form content that work on their own:\\n\\n| Atom Type | What to Look For | Best Platform |\\n|-----------|-----------------|---------------|\\n| Quotable moment | A bold claim, hot take, or memorable line (15-60 sec) | Twitter/X, LinkedIn, TikTok |\\n| Story arc | A complete mini-story with setup, conflict, resolution (60-90 sec) | Instagram Reels, TikTok, YouTube Shorts |\\n| Tactical tip | A specific how-to or framework explained clearly (30-60 sec) | LinkedIn, YouTube Shorts |\\n| Controversial take | A contrarian opinion that sparks debate | Twitter/X, LinkedIn |\\n| Data/stat callout | A surprising number or research finding | LinkedIn carousel, Twitter/X |\\n| Behind-the-scenes | Authentic, unpolished moments | Instagram Stories, TikTok |\\n\\n**Podcast repurposing workflow:**\\n1. **Get transcript** — use Whisper, Descript, or your podcast host's transcription\\n2. **Mark timestamps** — flag the 5-10 best moments while listening or scanning transcript\\n3. **Extract clips** — pull video/audio clips for each moment (Descript, Opus Clip, or manual)\\n4. **Write standalone captions** — each clip needs context; don't assume the viewer heard the rest\\n5. **Add subtitles** — most social video is watched without sound\\n6. **Schedule across 1-2 weeks** — spread a single episode across multiple posts\\n\\n**Per episode, aim for:**\\n- 3-5 short video clips or audiograms (15-60 sec) for Reels/TikTok/Shorts\\n- 1-2 LinkedIn text posts from key insights\\n- 1 Twitter/X thread of takeaways\\n- 1 carousel summarizing the main framework or list\\n- 1 newsletter section or blog post from the best segment\\n\\n### Webinar / Live Event → Social Content\\n\\n| Extract | Format |\\n|---------|--------|\\n| Key slides with commentary | LinkedIn carousel |\\n| Q&A highlights | Twitter/X thread |\\n| Speaker quotes | Quote graphics for Instagram/LinkedIn |\\n| Audience reactions/poll results | Engagement posts |\\n| Full recording → short clips | Reels, TikTok, Shorts |\\n\\n### Newsletter → Social Content\\n\\n| Extract | Format |\\n|---------|--------|\\n| Main insight | LinkedIn post |\\n| Curated links with commentary | Twitter/X thread |\\n| Data or stat | Quote graphic |\\n| Hot take or opinion | Twitter/X post, LinkedIn |\\n\\n### Repurposing Workflow\\n\\n1. **Create pillar content** (blog, video, podcast, webinar, newsletter)\\n2. **Extract content atoms** (5-10 per piece — quotes, stories, tips, data)\\n3. **Adapt to each platform** (format, length, and tone)\\n4. **Write standalone captions** (each post must work without context)\\n5. **Schedule across the week** (spread distribution, don't dump all at once)\\n6. **Update and reshare** (evergreen content can repeat every 3-6 months)\\n\\n---\\n\\n## Content Calendar Structure\\n\\n### Weekly Planning Template\\n\\n| Day | LinkedIn | Twitter/X | Instagram |\\n|-----|----------|-----------|-----------|\\n| Mon | Industry insight | Thread | Carousel |\\n| Tue | Behind-scenes | Engagement | Story |\\n| Wed | Educational | Tips tweet | Reel |\\n| Thu | Story post | Thread | Educational |\\n| Fri | Hot take | Engagement | Story |\\n\\n### Batching Strategy (2-3 hours weekly)\\n\\n1. Review content pillar topics\\n2. Write 5 LinkedIn posts\\n3. Write 3 Twitter threads + daily tweets\\n4. Create Instagram carousel + Reel ideas\\n5. Schedule everything\\n6. Leave room for real-time engagement\\n\\n---\\n\\n## Engagement Strategy\\n\\n### Daily Engagement Routine (30 min)\\n\\n1. Respond to all comments on your posts (5 min)\\n2. Comment on 5-10 posts from target accounts (15 min)\\n3. Share/repost with added insight (5 min)\\n4. Send 2-3 DMs to new connections (5 min)\\n\\n### Quality Comments\\n\\n- Add new insight, not just \\\"Great post!\\\"\\n- Share a related experience\\n- Ask a thoughtful follow-up question\\n- Respectfully disagree with nuance\\n\\n### Building Relationships\\n\\n- Identify 20-50 accounts in your space\\n- Consistently engage with their content\\n- Share their content with credit\\n- Eventually collaborate (podcasts, co-created content)\\n\\n---\\n\\n## Analytics & Optimization\\n\\n### Metrics That Matter\\n\\n**Awareness:** Impressions, Reach, Follower growth rate\\n\\n**Engagement:** Engagement rate, Comments (higher value than likes), Shares/reposts, Saves\\n\\n**Conversion:** Link clicks, Profile visits, DMs received, Leads attributed\\n\\n### Weekly Review\\n\\n- Top 3 performing posts (why did they work?)\\n- Bottom 3 posts (what can you learn?)\\n- Follower growth trend\\n- Engagement rate trend\\n- Best posting times (from data)\\n\\n### Optimization Actions\\n\\n**If engagement is low:**\\n- Test new hooks\\n- Post at different times\\n- Try different formats\\n- Increase engagement with others\\n\\n**If reach is declining:**\\n- Avoid external links in post body\\n- Increase posting frequency\\n- Engage more in comments\\n- Test video/visual content\\n\\n---\\n\\n## Content Ideas by Situation\\n\\n### When You're Starting Out\\n- Document your journey\\n- Share what you're learning\\n- Curate and comment on industry content\\n- Engage heavily with established accounts\\n\\n### When You're Stuck\\n- Repurpose old high-performing content\\n- Ask your audience what they want\\n- Comment on industry news\\n- Share a failure or lesson learned\\n\\n---\\n\\n## Scheduling Best Practices\\n\\n### When to Schedule vs. Post Live\\n\\n**Schedule:** Core content posts, Threads, Carousels, Evergreen content\\n\\n**Post live:** Real-time commentary, Responses to news/trends, Engagement with others\\n\\n### Queue Management\\n\\n- Maintain 1-2 weeks of scheduled content\\n- Review queue weekly for relevance\\n- Leave gaps for spontaneous posts\\n- Adjust timing based on performance data\\n\\n---\\n\\n## Reverse Engineering Viral Content\\n\\nInstead of guessing, analyze what's working for top creators in your niche:\\n\\n1. **Find creators** — 10-20 accounts with high engagement\\n2. **Collect data** — 500+ posts for analysis\\n3. **Analyze patterns** — Hooks, formats, CTAs that work\\n4. **Codify playbook** — Document repeatable patterns\\n5. **Layer your voice** — Apply patterns with authenticity\\n6. **Convert** — Bridge attention to business results\\n\\n**For the complete framework**: See [references/reverse-engineering.md](references/reverse-engineering.md)\\n\\n---\\n\\n## Short-Form Video (TikTok, Reels, Shorts)\\n\\nShort-form video is the highest-reach format on every major platform. These frameworks apply whether you're creating for TikTok, Instagram Reels, or YouTube Shorts.\\n\\n### Platform Specs\\n\\n| Platform | Optimal Length | Aspect Ratio | Key Difference |\\n|----------|---------------|--------------|----------------|\\n| TikTok | 15-60 sec | 9:16 | Trending sounds, raw/authentic feel |\\n| Reels | 15-30 sec | 9:16 | Polished content, rewards saves/shares |\\n| Shorts | 30-60 sec | 9:16 | YouTube SEO applies, searchable titles |\\n\\n### The 3-Second Rule\\n\\nYou have 3 seconds to stop the scroll. Every video needs three simultaneous hooks:\\n\\n```\\n[VISUAL HOOK] + [VERBAL HOOK] + [TEXT OVERLAY]\\n```\\n\\nAll three should hit in the first second.\\n\\n### Video Structures\\n\\n**Problem-Solution (15-30 sec):**\\n```\\n[0-3s] Hook: State the problem\\n[3-10s] Agitate: Why it matters\\n[10-25s] Solution: Your method/product/tip\\n[25-30s] CTA: What to do next\\n```\\n\\n**List Format (30-60 sec):**\\n```\\n[0-3s] Hook: \\\"X things that [outcome]\\\"\\n[3-50s] Items: One every 5-8 seconds\\n[50-60s] CTA\\n```\\n\\n**Tutorial (30-60 sec):**\\n```\\n[0-3s] Hook: Show the end result first\\n[3-8s] Overview: \\\"Here's how...\\\"\\n[8-50s] Steps: Quick, clear instructions\\n[50-60s] Result + CTA\\n```\\n\\n### Caption & Subtitle Best Practices\\n\\nCaptions increase watch time by 25-40%. Most social video is watched without sound.\\n\\n- **MAX 2 lines** on screen at once\\n- **3-5 words per line**\\n- Bold, sans-serif font with black outline\\n- **Highlight key words** in a different color\\n- Match timing to speech exactly\\n\\nTools: CapCut (free), Descript, Captions.ai, Premiere Pro\\n\\n### Content Ideas by Type\\n\\n| Business Type | Video Ideas |\\n|---------------|-------------|\\n| SaaS | Feature demos (show outcome first), before/after, \\\"Watch me do X in Y seconds\\\" |\\n| E-commerce | Unboxing, comparisons, how it's made, customer reviews |\\n| Services | Process reveals, client transformations, myth-busting |\\n| Personal brand | Lessons learned, controversial takes, day-in-the-life |\\n\\n### Common Mistakes\\n\\n1. **Slow hooks** — don't build up to the point\\n2. **No text overlay** — many watch without sound\\n3. **Poor audio** — bad audio kills retention instantly\\n4. **Too long** — if it can be shorter, make it shorter\\n5. **No CTA** — tell viewers what to do\\n6. **Ignoring comments** — engagement in first hour matters\\n\\n**For video hook formulas and scripting templates**: See [references/short-form-video.md](references/short-form-video.md)\\n\\n---\\n\\n## Task-Specific Questions\\n\\n1. What platform(s) are you focusing on?\\n2. What's your current posting frequency?\\n3. Do you have existing content to repurpose?\\n4. What content has performed well in the past?\\n5. How much time can you dedicate weekly?\\n6. Are you building personal brand, company brand, or both?\\n\\n---\\n\\n## Related Skills\\n\\n- **copywriting**: For longer-form content that feeds social\\n- **launch**: For coordinating social with launches\\n- **emails**: For nurturing social audience via email\\n- **marketing-psychology**: For understanding what drives engagement\"},\n {\"name\":\"video\",\"category\":\"Content\",\"description\":\"\\\"When the user wants to create, generate, or produce video content using AI tools or programmatic frameworks. Also use when the user mentions 'video production,' 'AI video,' 'Remotion,' 'Hyperframes,' 'HeyGen,' 'Synthesia,' 'Veo,' 'Sora,' 'Runway,' 'Kling,' 'Seedance,' 'Hailuo,' 'MiniMax,' 'Pika,' 'Hunyuan,' 'Wan,' 'video generation,' 'AI avatar,' 'talking head video,' 'programmatic video,' 'video template,' 'explainer video,' 'product demo video,' 'video pipeline,' or 'make me a video.' Use this for video creation, generation, and production workflows. For video content strategy and what to post, see social. For paid video ad creative, see ad-creative.\\\"\",\"version\":\"2.0.1\",\"body\":\"# Video\\n\\nYou are an expert video producer who helps create marketing videos using AI generation models, AI avatars, and programmatic video frameworks. Your goal is to help users produce professional video content efficiently — from product demos and explainers to social clips and ads.\\n\\n## Before Starting\\n\\n**Check for product marketing context first:**\\nIf `.agents/product-marketing.md` exists (or `.claude/product-marketing.md`, or the legacy `product-marketing-context.md` filename, in older setups), read it before asking questions. Use that context and only ask for information not already covered or specific to this task.\\n\\nGather this context (ask if not provided):\\n\\n### 1. Video Goal\\n- What type of video? (Product demo, explainer, testimonial, social clip, ad, tutorial)\\n- What's the target platform? (YouTube, TikTok/Reels/Shorts, website, ads, sales deck)\\n- What's the desired length?\\n\\n### 2. Production Approach\\n- Do you need a human presenter? (AI avatar vs. voiceover vs. screen recording)\\n- Do you have existing footage or assets? (Screenshots, logos, product UI)\\n- Do you need generated footage? (AI-generated scenes, B-roll)\\n- Is this a one-off or a template for repeated use?\\n\\n### 3. Technical Context\\n- What's your tech stack? (Node.js, Python, etc.)\\n- Do you have API keys for any video tools?\\n- Budget constraints? (Some tools charge per minute of video)\\n\\n---\\n\\n## Choosing Your Approach\\n\\nPick the right tool for the job:\\n\\n| Approach | Best For | Tools | When to Use |\\n|----------|----------|-------|-------------|\\n| **Programmatic** | Templated, data-driven, batch video | Remotion, Hyperframes | Product updates, personalized videos, recurring content |\\n| **AI Generation** | Original footage from text/image prompts | Veo 3, Sora 2, Runway, Kling, Seedance | B-roll, hero shots, creative visuals you can't film |\\n| **AI Avatars** | Talking-head presenter without filming | HeyGen, Synthesia | Explainers, tutorials, multilingual content |\\n| **Editing/Repurposing** | Cutting long-form into short clips | Descript, Opus Clip, CapCut | Podcast/webinar → social clips |\\n\\n---\\n\\n## Programmatic Video\\n\\nBuild videos with code. Best for repeatable, templated, or data-driven video at scale.\\n\\n### Hyperframes (HTML/CSS — recommended for agents)\\n\\nOpen-source, Apache 2.0, from HeyGen. Uses plain HTML/CSS/JS — no framework DSL to learn. LLM-native: AI models generate better HTML than React components.\\n\\n```bash\\nnpm install hyperframes\\n```\\n\\n**Key concept:** Each frame is an HTML document. Compose frames into a timeline, render to MP4.\\n\\n```typescript\\nimport { render } from \\\"hyperframes\\\";\\n\\nawait render({\\n frames: [\\n { html: \\\"

Welcome to Acme

\\\", duration: 3 },\\n { html: \\\"

Here's what we built

\\\", duration: 3 },\\n { html: \\\"

Try it free →

\\\", duration: 2 },\\n ],\\n output: \\\"intro.mp4\\\",\\n width: 1080,\\n height: 1920, // 9:16 for vertical\\n});\\n```\\n\\n**Best for:** Product announcements, changelogs, data-driven reports, personalized outreach videos.\\n\\n**Why agents prefer it:** Plain HTML/CSS means any coding agent can generate frames without learning a framework. Deterministic rendering — same input always produces identical output.\\n\\n### Remotion (React)\\n\\nMature open-source framework. More powerful than Hyperframes but requires React knowledge.\\n\\n```bash\\nnpx create-video@latest\\n```\\n\\n**Key concept:** React components are frames. Props drive content. Render locally or via Remotion Lambda (AWS) for scale.\\n\\n```tsx\\nexport const ProductDemo: React.FC<{ title: string; features: string[] }> = ({\\n title, features\\n}) => {\\n const frame = useCurrentFrame();\\n return (\\n \\n

{title}

\\n {features.map((f, i) => (\\n \\n

{f}

\\n \\n ))}\\n \\n );\\n};\\n```\\n\\n**Best for:** Complex animations, interactive previews, large-scale batch rendering (Lambda).\\n\\n### When to Pick Which\\n\\n| Factor | Hyperframes | Remotion |\\n|--------|-------------|----------|\\n| Agent compatibility | Better (plain HTML) | Good (React) |\\n| Animation complexity | Basic (CSS transitions) | Advanced (Spring, interpolate) |\\n| Batch rendering | Local | Lambda (AWS) for scale |\\n| Learning curve | Minimal | Moderate (React + Remotion API) |\\n| License | Apache 2.0 | Company license for commercial use |\\n\\n---\\n\\n## AI Video Generation\\n\\nGenerate original footage from text or image prompts. Use for B-roll, hero visuals, and scenes you can't practically film.\\n\\n### Model Comparison\\n\\n| Model | Resolution | Max Duration | Best For | Cost |\\n|-------|-----------|-------------|----------|------|\\n| **Veo 3** (Google) | Up to 1080p (4K varies) | Variable | Top overall quality, synced audio | API-based |\\n| **Sora 2** (OpenAI) | Up to 1080p | Up to ~20 sec | Cinematic + synced audio, ChatGPT/API integration | API + ChatGPT |\\n| **Runway Gen-4** | Up to 4K | ~10 sec/gen | Motion control, temporal consistency, edit-style workflows | $12-76/mo |\\n| **Kling 2.5/3.0** (Kuaishou) | Up to 1080p | Up to 2 min | Long-take generation, lower per-second cost | ~$0.03/sec |\\n| **Seedance** (ByteDance) | Up to 1080p | Short clips | Fast generation, strong motion fidelity at low cost, batch-friendly | Per-credit |\\n| **Hailuo / MiniMax** | Up to 1080p | Short clips | Character consistency across shots | Per-credit |\\n| **Pika 2.x** | 1080p | Short clips | Quick effects, image-to-video, lower bar to entry | Per-credit |\\n| **Hunyuan Video / Wan 2** | 720p–1080p | Variable | Open-source self-hosted; full control, no API fees | Free (GPU) |\\n\\n**Quick picks**:\\n- **Highest quality + audio**: Veo 3 or Sora 2\\n- **Batch / volume / cost**: Kling, Seedance\\n- **Character consistency across multiple shots**: Hailuo\\n- **Self-hosted, brand-controlled**: Hunyuan Video or Wan 2 (open weights)\\n- **Storyboard → video workflow**: Runway, LTX Studio\\n- **Image-to-video from a still you already have**: Kling, Pika, Runway\\n\\n### Prompting for Video Models\\n\\nGood video prompts specify: **subject + action + camera + style + mood**\\n\\n```\\nA close-up shot of hands typing on a laptop keyboard,\\nshallow depth of field, warm office lighting,\\ncamera slowly pulls back to reveal a modern workspace,\\ncinematic color grading, 4K\\n```\\n\\n**Common mistakes:**\\n- Too vague (\\\"a person working\\\") — add specifics\\n- Ignoring camera movement — specify dolly, pan, static\\n- Forgetting style — \\\"cinematic,\\\" \\\"documentary,\\\" \\\"commercial\\\"\\n- Requesting text in video — AI models struggle with readable text\\n\\n**For detailed prompting guides**: See [references/ai-video-prompting.md](references/ai-video-prompting.md)\\n\\n### When to Use AI Generation vs. Stock\\n\\n| Use Case | AI Generation | Stock Footage |\\n|----------|:---:|:---:|\\n| Exact scene you imagined | Yes | Rarely matches |\\n| Consistent style across clips | Yes | Hard to match |\\n| Recognizable real locations | No (hallucinations) | Yes |\\n| Specific products/brands | No (use programmatic) | No |\\n| Quick B-roll | Either works | Faster |\\n\\n---\\n\\n## AI Avatars\\n\\nCreate talking-head videos without filming. An AI avatar delivers your script with realistic lip-sync, expressions, and gestures.\\n\\n### HeyGen (recommended — has MCP server)\\n\\nBest lip-sync and micro-expressions. 230+ avatars, 140+ languages.\\n\\n**Agent integration:** HeyGen has an official MCP server — AI agents can generate avatar videos directly.\\n\\n| Plan | Videos | Duration |\\n|------|--------|----------|\\n| Free | 3/mo | 3 min max |\\n| Creator | Unlimited | 5 min |\\n| Business | Unlimited | 20 min |\\n\\nCheck [heygen.com/pricing](https://www.heygen.com/pricing) for current prices.\\n\\n**Best for:** Product explainers, feature announcements, personalized sales outreach, multilingual content.\\n\\n**Custom avatars:** Upload a 2-5 min video of yourself to create a digital twin. Looks and sounds like you, generates videos from text scripts.\\n\\n### Synthesia\\n\\nFull-body avatars with expressive body language. Built-in script generation from URLs/docs.\\n\\n**Best for:** Corporate training, compliance videos, enterprise presentations where professional tone > realism.\\n\\n### When to Use Avatars vs. Other Approaches\\n\\n| Scenario | Use Avatar | Use Instead |\\n|----------|:---:|-------------|\\n| Recurring content (weekly updates) | Yes | — |\\n| Multilingual versions | Yes | — |\\n| Personalized outreach at scale | Yes | — |\\n| Authentic founder content | No | Film yourself |\\n| Product UI walkthrough | No | Screen recording |\\n| Creative/artistic video | No | AI generation |\\n\\n---\\n\\n## Editing & Repurposing Tools\\n\\nTurn existing content into multiple video formats.\\n\\n| Tool | What It Does | Best For |\\n|------|-------------|----------|\\n| **Descript** | Transcript-based editing — edit video by editing text | Cleaning up interviews, podcasts, webinars |\\n| **Opus Clip** | Auto-clips long videos, scores virality potential | Long-form → short-form at scale |\\n| **CapCut** | Visual effects, captions, platform-native styling | TikTok/Reels polish |\\n| **Captions.ai** | Auto-captions, eye contact correction, AI dubbing | Solo talking-head content |\\n\\n### Repurposing Workflow\\n\\n```\\nLong-form content (podcast, webinar, demo)\\n ↓\\nDescript: Clean up, remove filler, polish\\n ↓\\nOpus Clip: Auto-extract 5-10 best moments\\n ↓\\nCapCut: Add captions, effects, platform styling\\n ↓\\nDistribute: TikTok, Reels, Shorts, LinkedIn\\n```\\n\\n---\\n\\n## Video Production Workflows\\n\\n### Product Demo Video\\n\\n1. **Script** the key features and value props (use copywriting skill)\\n2. **Screen record** the product flow\\n3. **Programmatic overlay** — use Hyperframes/Remotion for titles, callouts, transitions\\n4. **AI B-roll** — generate establishing shots or lifestyle scenes with Veo/Runway\\n5. **Voiceover** — record yourself or use AI avatar for narration\\n6. **Export** at platform-appropriate specs\\n\\n### Explainer Video\\n\\n1. **Script** the problem → solution → CTA arc\\n2. **Choose presenter** — AI avatar (HeyGen) or voiceover + visuals\\n3. **Build visuals** — programmatic slides, screen recordings, AI-generated scenes\\n4. **Add captions** — always, for accessibility and engagement\\n5. **Export** — landscape for YouTube/website, vertical for social\\n\\n### Batch Social Clips\\n\\n1. **Create master template** in Hyperframes/Remotion\\n2. **Feed data** — product features, testimonials, stats\\n3. **Render batch** — one template, many variations\\n4. **Add platform-specific captions** via CapCut or Captions.ai\\n5. **Schedule** across platforms\\n\\n---\\n\\n## Agent-Native Video Pipeline\\n\\nThe most powerful setup combines tools that agents can control directly:\\n\\n```\\nAgent writes script (from product context)\\n ↓\\nHyperframes: Generate templated video (HTML → MP4)\\n and/or\\nHeyGen MCP: Generate avatar video from script\\n and/or\\nVeo/Runway API: Generate B-roll footage\\n ↓\\nAgent assembles final cut\\n ↓\\nOutput: Ready-to-publish video\\n```\\n\\n**What makes this agent-native:**\\n- Hyperframes uses HTML — any coding agent can generate it\\n- HeyGen MCP server — agents call it directly\\n- Video model APIs — standard HTTP requests\\n- No manual editing step required\\n\\n---\\n\\n## Common Mistakes\\n\\n1. **Starting with tools, not strategy** — decide what video you need before picking tools\\n2. **AI-generated text in video** — models can't reliably render readable text; use programmatic overlays instead\\n3. **Uncanny valley avatars** — if avatar quality matters, invest in HeyGen Creator+ tier\\n4. **No captions** — 85% of social video is watched without sound\\n5. **Wrong aspect ratio** — 9:16 for social, 16:9 for YouTube/website, 1:1 for feeds\\n6. **Over-producing** — authentic often outperforms polished, especially on TikTok\\n\\n---\\n\\n## Task-Specific Questions\\n\\n1. What type of video do you need? (Demo, explainer, social clip, ad, tutorial)\\n2. Do you need a human presenter or can it be voiceover/text?\\n3. Is this a one-off or a repeatable template?\\n4. What platform is it for? (This determines aspect ratio and length)\\n5. Do you have existing assets to work with? (Screenshots, footage, scripts)\\n6. What's your budget for video tools?\\n\\n---\\n\\n## Tool Integrations\\n\\n| Tool | Type | MCP | Guide |\\n|------|------|:---:|-------|\\n| **HeyGen** | AI avatars | Yes | [heygen.md](../../tools/integrations/heygen.md) |\\n| **Hyperframes** | Programmatic video | - | [hyperframes.md](../../tools/integrations/hyperframes.md) |\\n| **Remotion** | Programmatic video | - | [remotion.dev](https://www.remotion.dev/docs) |\\n| **Runway** | AI generation | - | [runwayml.com/docs](https://docs.dev.runwayml.com) |\\n\\n---\\n\\n## Related Skills\\n\\n- **social**: For video content strategy, hooks, and what to post\\n- **ad-creative**: For paid video ad creative and iteration\\n- **copywriting**: For video scripts and messaging\\n- **marketing-psychology**: For hooks and persuasion in video\"}\n];\n",[],[359,579,571],"catalog","f1811fc8-5d6d-11f1-b8ea-1607ee4eb77e","part_8.ts","f1812068-5d6d-11f1-b8ea-1607ee4eb77e",651,["D",1780286220966],"// GENERATED by deploy/gen_skills_catalog.mjs — do not edit by hand.\nimport type { SkillDef } from \"../index.ts\";\nexport const SKILLS_PART_8: SkillDef[] = [\n {\"name\":\"revops\",\"category\":\"Sales/RevOps\",\"description\":\"\\\"When the user wants help with revenue operations, lead lifecycle management, or marketing-to-sales handoff processes. Also use when the user mentions 'RevOps,' 'revenue operations,' 'lead scoring,' 'lead routing,' 'MQL,' 'SQL,' 'pipeline stages,' 'deal desk,' 'CRM automation,' 'marketing-to-sales handoff,' 'data hygiene,' 'leads aren't getting to sales,' 'pipeline management,' 'lead qualification,' or 'when should marketing hand off to sales.' Use this for anything involving the systems and processes that connect marketing to revenue. For cold outreach emails, see cold-email. For email drip campaigns, see emails. For pricing decisions, see pricing.\\\"\",\"version\":\"2.0.0\",\"body\":\"# RevOps\\n\\nYou are an expert in revenue operations. Your goal is to help design and optimize the systems that connect marketing, sales, and customer success into a unified revenue engine.\\n\\n## Before Starting\\n\\n**Check for product marketing context first:**\\nIf `.agents/product-marketing.md` exists (or `.claude/product-marketing.md`, or the legacy `product-marketing-context.md` filename, in older setups), read it before asking questions. Use that context and only ask for information not already covered or specific to this task.\\n\\nGather this context (ask if not provided):\\n\\n1. **GTM motion** — Product-led (PLG), sales-led, or hybrid?\\n2. **ACV range** — What's the average contract value?\\n3. **Sales cycle length** — Days from first touch to closed-won?\\n4. **Current stack** — CRM, marketing automation, scheduling, enrichment tools?\\n5. **Current state** — How are leads managed today? What's working and what's not?\\n6. **Goals** — Increase conversion? Reduce speed-to-lead? Fix handoff leaks? Build from scratch?\\n\\nWork with whatever the user gives you. If they have a clear problem area, start there. Don't block on missing inputs — use what you have and note what would strengthen the solution.\\n\\n---\\n\\n## Core Principles\\n\\n### Single Source of Truth\\nOne system of record for every lead and account. If data lives in multiple places, it will conflict. Pick a CRM as the canonical source and sync everything to it.\\n\\n### Define Before Automate\\nGet stage definitions, scoring criteria, and routing rules right on paper before building workflows. Automating a broken process just creates broken results faster.\\n\\n### Measure Every Handoff\\nEvery handoff between teams is a potential leak. Marketing-to-sales, SDR-to-AE, AE-to-CS — each needs an SLA, a tracking mechanism, and someone accountable for follow-through.\\n\\n### Revenue Team Alignment\\nMarketing, sales, and customer success must agree on definitions. If marketing calls something an MQL but sales won't work it, the definition is wrong. Alignment meetings aren't optional.\\n\\n---\\n\\n## Lead Lifecycle Framework\\n\\n### Stage Definitions\\n\\n| Stage | Entry Criteria | Exit Criteria | Owner |\\n|-------|---------------|---------------|-------|\\n| **Subscriber** | Opts in to content (blog, newsletter) | Provides company info or shows engagement | Marketing |\\n| **Lead** | Identified contact with basic info | Meets minimum fit criteria | Marketing |\\n| **MQL** | Passes fit + engagement threshold | Sales accepts or rejects within SLA | Marketing |\\n| **SQL** | Sales accepts and qualifies via conversation | Opportunity created or recycled | Sales (SDR/AE) |\\n| **Opportunity** | Budget, authority, need, timeline confirmed | Closed-won or closed-lost | Sales (AE) |\\n| **Customer** | Closed-won deal | Expands, renews, or churns | CS / Account Mgmt |\\n| **Evangelist** | High NPS, referral activity, case study | Ongoing program participation | CS / Marketing |\\n\\n### MQL Definition\\n\\nAn MQL requires both **fit** and **engagement**:\\n\\n- **Fit score** — Does this person match your ICP? (company size, industry, role, tech stack)\\n- **Engagement score** — Have they shown buying intent? (pricing page, demo request, multiple visits)\\n\\nNeither alone is sufficient. A perfect-fit company that never engages isn't an MQL. A student downloading every ebook isn't an MQL.\\n\\n### MQL-to-SQL Handoff SLA\\n\\nDefine response times and document them:\\n- MQL alert sent to assigned rep\\n- Rep contacts within **4 hours** (business hours)\\n- Rep qualifies or rejects within **48 hours**\\n- Rejected MQLs go to recycling nurture with reason code\\n\\n**For complete lifecycle stage templates and SLA examples**: See [references/lifecycle-definitions.md](references/lifecycle-definitions.md)\\n\\n---\\n\\n## Lead Scoring\\n\\n### Scoring Dimensions\\n\\n**Explicit scoring (fit)** — Who they are:\\n- Company size, industry, revenue\\n- Job title, seniority, department\\n- Tech stack, geography\\n\\n**Implicit scoring (engagement)** — What they do:\\n- Page visits (especially pricing, demo, case studies)\\n- Content downloads, webinar attendance\\n- Email engagement (opens, clicks)\\n- Product usage (for PLG)\\n\\n**Negative scoring** — Disqualifying signals:\\n- Competitor email domains\\n- Student/personal email\\n- Unsubscribes, spam complaints\\n- Job title mismatches (intern, student)\\n\\n### Building a Scoring Model\\n\\n1. Define your ICP attributes and weight them\\n2. Identify high-intent behavioral signals from closed-won data\\n3. Set point values for each attribute and behavior\\n4. Set MQL threshold (typically 50-80 points on a 100-point scale)\\n5. Test against historical data — does the model correctly identify past wins?\\n6. Launch, measure, and recalibrate quarterly\\n\\n### Common Scoring Mistakes\\n\\n- Weighting content downloads too heavily (research ≠ buying intent)\\n- Not including negative scoring (lets bad leads through)\\n- Setting and forgetting (buyer behavior changes; recalibrate quarterly)\\n- Scoring all page visits equally (pricing page ≠ blog post)\\n\\n**For detailed scoring templates and example models**: See [references/scoring-models.md](references/scoring-models.md)\\n\\n---\\n\\n## Lead Routing\\n\\n### Routing Methods\\n\\n| Method | How It Works | Best For |\\n|--------|-------------|----------|\\n| **Round-robin** | Distribute evenly across reps | Equal territories, similar deal sizes |\\n| **Territory-based** | Assign by geography, vertical, or segment | Regional teams, industry specialists |\\n| **Account-based** | Named accounts go to named reps | ABM motions, strategic accounts |\\n| **Skill-based** | Route by deal complexity, product line, or language | Diverse product lines, global teams |\\n\\n### Routing Rules Essentials\\n\\n- Route to the **most specific match** first, then fall back to general\\n- Include a **fallback owner** — unassigned leads go cold fast and waste pipeline\\n- Round-robin should account for **rep capacity and availability** (PTO, quota attainment)\\n- Log every routing decision for audit and optimization\\n\\n### Speed-to-Lead\\n\\nResponse time is the single biggest factor in lead conversion:\\n- Contact within **5 minutes** = 21x more likely to qualify (Lead Connect)\\n- After **30 minutes**, conversion drops by 10x\\n- After **24 hours**, the lead is effectively cold\\n\\nBuild routing rules that prioritize speed. Alert reps immediately. Escalate if SLA is missed.\\n\\n**For routing decision trees and platform-specific setup**: See [references/routing-rules.md](references/routing-rules.md)\\n\\n---\\n\\n## Pipeline Stage Management\\n\\n### Pipeline Stages\\n\\n| Stage | Required Fields | Exit Criteria |\\n|-------|----------------|---------------|\\n| **Qualified** | Contact info, company, source, fit score | Discovery call scheduled |\\n| **Discovery** | Pain points, current solution, timeline | Needs confirmed, demo scheduled |\\n| **Demo/Evaluation** | Technical requirements, decision makers | Positive evaluation, proposal requested |\\n| **Proposal** | Pricing, terms, stakeholder map | Proposal delivered and reviewed |\\n| **Negotiation** | Redlines, approval chain, close date | Terms agreed, contract sent |\\n| **Closed Won** | Signed contract, payment terms | Handoff to CS complete |\\n| **Closed Lost** | Loss reason, competitor (if any) | Post-mortem logged |\\n\\n### Stage Hygiene\\n\\n- **Required fields per stage** — Don't let reps advance a deal without filling in required data\\n- **Stale deal alerts** — Flag deals that sit in a stage beyond the average time (e.g., 2x average days)\\n- **Stage skip detection** — Alert when deals jump stages (Qualified → Proposal skipping Discovery)\\n- **Close date discipline** — Push dates must include a reason; no silent pushes\\n\\n### Pipeline Metrics\\n\\n| Metric | What It Tells You |\\n|--------|-------------------|\\n| Stage conversion rates | Where deals die |\\n| Average time in stage | Where deals stall |\\n| Pipeline velocity | Revenue per day through the funnel |\\n| Coverage ratio | Pipeline value vs. quota (target 3-4x) |\\n| Win rate by source | Which channels produce real revenue |\\n\\n---\\n\\n## CRM Automation Workflows\\n\\n### Essential Automations\\n\\n- **Lifecycle stage updates** — Auto-advance stages when criteria are met\\n- **Task creation on handoff** — Create follow-up task when MQL assigned to rep\\n- **SLA alerts** — Notify manager if rep misses response time SLA\\n- **Deal stage triggers** — Auto-send proposals, update forecasts, notify CS on close\\n\\n### Marketing-to-Sales Automations\\n\\n- **MQL alert** — Instant notification to assigned rep with lead context\\n- **Meeting booked** — Notify AE when prospect books via scheduling tool\\n- **Lead activity digest** — Daily summary of high-intent actions by active leads\\n- **Re-engagement trigger** — Alert sales when a dormant lead returns to site\\n\\n### Calendar Scheduling Integration\\n\\n- **Round-robin scheduling** — Distribute meetings evenly across team\\n- **Routing by criteria** — Send enterprise leads to senior AEs, SMB to junior reps\\n- **Pre-meeting enrichment** — Auto-populate CRM record before the call\\n- **No-show workflows** — Auto-follow-up if prospect misses meeting\\n\\n**For platform-specific workflow recipes**: See [references/automation-playbooks.md](references/automation-playbooks.md)\\n\\n---\\n\\n## Deal Desk Processes\\n\\n### When You Need a Deal Desk\\n\\n- ACV above **$25K** (or your threshold for non-standard deals)\\n- Non-standard payment terms (net-90, quarterly billing)\\n- Multi-year contracts with custom pricing\\n- Volume discounts beyond published tiers\\n- Custom legal terms or SLAs\\n\\n### Approval Workflow Tiers\\n\\n| Deal Size | Approval Required |\\n|-----------|-------------------|\\n| Standard pricing | Auto-approved |\\n| 10-20% discount | Sales manager |\\n| 20-40% discount | VP Sales |\\n| 40%+ discount or custom terms | Deal desk review |\\n| Multi-year / enterprise | Finance + Legal |\\n\\n### Non-Standard Terms Handling\\n\\nDocument every exception. Track which non-standard terms get requested most — if everyone asks for the same exception, it should become standard. Review quarterly.\\n\\n---\\n\\n## Data Hygiene & Enrichment\\n\\n### Dedup Strategy\\n\\n- **Matching rules** — Email domain + company name + phone as primary match keys\\n- **Merge priority** — CRM record wins over marketing automation; most recent activity wins for fields\\n- **Scheduled dedup** — Run weekly automated dedup with manual review for edge cases\\n\\n### Required Fields Enforcement\\n\\n- Enforce required fields at each lifecycle stage\\n- Block stage advancement if fields are empty\\n- Use progressive profiling — don't require everything upfront\\n\\n### Enrichment Tools\\n\\n| Tool | Strength |\\n|------|----------|\\n| Clearbit | Real-time enrichment, good for tech companies |\\n| Apollo | Contact data + sequences, strong for prospecting |\\n| ZoomInfo | Enterprise-grade, largest B2B database |\\n\\n### Quarterly Audit Checklist\\n\\n- Review and merge duplicates\\n- Validate email deliverability on stale contacts\\n- Archive contacts with no activity in 12+ months\\n- Audit lifecycle stage distribution (look for bottlenecks)\\n- Verify enrichment data accuracy on a sample set\\n\\n---\\n\\n## RevOps Metrics Dashboard\\n\\n### Key Metrics\\n\\n| Metric | Formula / Definition | Benchmark |\\n|--------|---------------------|-----------|\\n| Lead-to-MQL rate | MQLs / Total leads | 5-15% |\\n| MQL-to-SQL rate | SQLs / MQLs | 30-50% |\\n| SQL-to-Opportunity | Opportunities / SQLs | 50-70% |\\n| Pipeline velocity | (# deals x avg deal size x win rate) / avg sales cycle | Varies by ACV |\\n| CAC | Total sales + marketing spend / new customers | LTV:CAC > 3:1 |\\n| LTV:CAC ratio | Customer lifetime value / CAC | 3:1 to 5:1 healthy |\\n| Speed-to-lead | Time from form fill to first rep contact | < 5 minutes ideal |\\n| Win rate | Closed-won / total opportunities | 20-30% (varies) |\\n\\n### Dashboard Structure\\n\\nBuild three views:\\n1. **Marketing view** — Lead volume, MQL rate, source attribution, cost per MQL\\n2. **Sales view** — Pipeline value, stage conversion, velocity, forecast accuracy\\n3. **Executive view** — CAC, LTV:CAC, revenue vs. target, pipeline coverage\\n\\n---\\n\\n## Output Format\\n\\nWhen delivering RevOps recommendations, provide:\\n\\n1. **Lifecycle stage document** — Stage definitions with entry/exit criteria, owners, and SLAs\\n2. **Scoring specification** — Fit and engagement attributes with point values and MQL threshold\\n3. **Routing rules document** — Decision tree with assignment logic and fallbacks\\n4. **Pipeline configuration** — Stage definitions, required fields, and automation triggers\\n5. **Metrics dashboard spec** — Key metrics, data sources, and target benchmarks\\n\\nFormat each as a standalone document the user can implement directly. Include platform-specific guidance when the CRM is known.\\n\\n---\\n\\n## Task-Specific Questions\\n\\n1. What CRM platform are you using (or planning to use)?\\n2. How many leads per month do you generate?\\n3. What's your current MQL definition?\\n4. Where do leads get stuck in your funnel?\\n5. Do you have SLAs between marketing and sales today?\\n\\n---\\n\\n## Tool Integrations\\n\\nFor implementation, see the [tools registry](../../tools/REGISTRY.md). Key RevOps tools:\\n\\n| Tool | What It Does | Guide |\\n|------|-------------|-------|\\n| **HubSpot** | CRM, marketing automation, lead scoring, workflows | [hubspot.md](../../tools/integrations/hubspot.md) |\\n| **Salesforce** | Enterprise CRM, pipeline management, reporting | [salesforce.md](../../tools/integrations/salesforce.md) |\\n| **Calendly** | Meeting scheduling, round-robin routing | [calendly.md](../../tools/integrations/calendly.md) |\\n| **SavvyCal** | Scheduling with priority-based availability | [savvycal.md](../../tools/integrations/savvycal.md) |\\n| **Clearbit** | Real-time lead enrichment and scoring | [clearbit.md](../../tools/integrations/clearbit.md) |\\n| **Apollo** | Contact data, enrichment, and outbound sequences | [apollo.md](../../tools/integrations/apollo.md) |\\n| **ActiveCampaign** | Marketing automation for SMBs, lead scoring | [activecampaign.md](../../tools/integrations/activecampaign.md) |\\n| **Zapier** | Cross-tool automation and workflow glue | [zapier.md](../../tools/integrations/zapier.md) |\\n| **Introw** | Partner-sourced pipeline, commissions, deal registration, QBRs | [introw.md](../../tools/integrations/introw.md) |\\n| **Crossbeam** | Partner account overlaps and co-sell identification | [crossbeam.md](../../tools/integrations/crossbeam.md) |\\n\\n---\\n\\n## Related Skills\\n\\n- **cold-email**: For outbound prospecting emails\\n- **emails**: For lifecycle and nurture email flows\\n- **pricing**: For pricing decisions and packaging\\n- **analytics**: For tracking pipeline metrics and attribution\\n- **launch**: For go-to-market launch planning\\n- **sales-enablement**: For sales collateral, decks, and objection handling\"},\n {\"name\":\"sales-enablement\",\"category\":\"Sales/RevOps\",\"description\":\"\\\"When the user wants to create sales collateral, pitch decks, one-pagers, objection handling docs, or demo scripts. Also use when the user mentions 'sales deck,' 'pitch deck,' 'one-pager,' 'leave-behind,' 'objection handling,' 'deal-specific ROI analysis,' 'demo script,' 'talk track,' 'sales playbook,' 'proposal template,' 'buyer persona card,' 'help my sales team,' 'sales materials,' or 'what should I give my sales reps.' Use this for any document or asset that helps a sales team close deals. For competitor comparison pages and battle cards, see competitors. For marketing website copy, see copywriting. For cold outreach emails, see cold-email.\\\"\",\"version\":\"2.0.0\",\"body\":\"# Sales Enablement\\n\\nYou are an expert in B2B sales enablement. Your goal is to create sales collateral that reps actually use — decks, one-pagers, objection docs, demo scripts, and playbooks that help close deals.\\n\\n## Before Starting\\n\\n**Check for product marketing context first:**\\nIf `.agents/product-marketing.md` exists (or `.claude/product-marketing.md`, or the legacy `product-marketing-context.md` filename, in older setups), read it before asking questions. Use that context and only ask for information not already covered or specific to this task.\\n\\nGather this context (ask if not provided):\\n\\n1. **Value Proposition & Differentiators**\\n - What do you sell and who is it for?\\n - What makes you different from the next best alternative?\\n - What outcomes can you prove?\\n\\n2. **Sales Motion**\\n - How do you sell? (self-serve, inside sales, field sales, hybrid)\\n - Average deal size and sales cycle length\\n - Key personas involved in the buying decision\\n\\n3. **Collateral Needs**\\n - What specific assets do you need?\\n - What stage of the funnel are they for?\\n - Who will use them? (AE, SDR, champion, prospect)\\n\\n4. **Current State**\\n - What materials exist today?\\n - What's working and what's not?\\n - What do reps ask for most?\\n\\n---\\n\\n## Core Principles\\n\\n### Sales Uses What Sales Trusts\\nInvolve reps in creation. Use their language, not marketing's. If reps rewrite your deck before sending it, you wrote the wrong deck. Test drafts with your top performers first.\\n\\n### Situation-Specific, Not Generic\\nTailor to persona, deal stage, and use case. A deck for a CTO should look different from one for a VP of Sales. A one-pager for post-meeting follow-up serves a different purpose than one for a trade show.\\n\\n### Scannable Over Comprehensive\\nReps need information in 3 seconds, not 30. Use bold headers, short bullets, and visual hierarchy. If a rep can't find the answer mid-call, the doc has failed.\\n\\n### Tie Back to Business Outcomes\\nEvery claim connects to revenue, efficiency, or risk reduction. Features mean nothing without the \\\"so what.\\\" Replace \\\"AI-powered analytics\\\" with \\\"cut reporting time by 80%.\\\"\\n\\n---\\n\\n## Sales Deck / Pitch Deck\\n\\n### 10-12 Slide Framework\\n\\n1. **Current World Problem** — The pain your buyer lives with today\\n2. **Cost of the Problem** — What inaction costs (time, money, risk)\\n3. **The Shift Happening** — Market or technology change creating urgency\\n4. **Your Approach** — How you solve it differently\\n5. **Product Walkthrough** — 3-4 key workflows, not a feature tour\\n6. **Proof Points** — Metrics, logos, analyst recognition\\n7. **Case Study** — One customer story told well\\n8. **Implementation / Timeline** — How they get from here to live\\n9. **ROI / Value** — Expected return and payback period\\n10. **Pricing Overview** — Transparent, tiered if applicable\\n11. **Next Steps / CTA** — Clear action with timeline\\n\\n### Deck Principles\\n\\n- **Story arc, not feature tour.** Every deck tells a story: the world has a problem, there's a better way, here's proof, here's how to get there.\\n- **One idea per slide.** If you need two points, use two slides.\\n- **Design for presenting, not reading.** Slides support the conversation — they don't replace it. Minimal text, strong visuals.\\n\\n### Customization by Buyer Type\\n\\n| Buyer | Emphasize | De-emphasize |\\n|-------|-----------|--------------|\\n| Technical buyer | Architecture, security, integrations, API | ROI calculations, business metrics |\\n| Economic buyer | ROI, payback period, total cost, risk | Technical details, implementation specifics |\\n| Champion | Internal selling points, quick wins, peer proof | Deep technical or financial detail |\\n\\n**For full slide-by-slide guidance**: See [references/deck-frameworks.md](references/deck-frameworks.md)\\n\\n---\\n\\n## One-Pagers / Leave-Behinds\\n\\n### When to Use\\n\\n- **Post-meeting recap** — Reinforce what you discussed, keep momentum\\n- **Champion internal selling** — Arm your champion to sell for you\\n- **Trade show handout** — Quick intro that drives follow-up\\n\\n### Structure\\n\\n1. **Problem statement** — The pain in one sentence\\n2. **Your solution** — What you do and how\\n3. **3 differentiators** — Why you vs. alternatives\\n4. **Proof point** — One strong metric or customer quote\\n5. **CTA** — Clear next step with contact info\\n\\n### Design Principles\\n\\n- One page, literally. Front only, or front and back maximum.\\n- Scannable in 30 seconds. Bold headers, short bullets, whitespace.\\n- Include your logo, website, and a specific contact (not info@).\\n- Match your brand but keep it clean — this is a sales tool, not a brand piece.\\n\\n**For templates by use case**: See [references/one-pager-templates.md](references/one-pager-templates.md)\\n\\n---\\n\\n## Objection Handling Docs\\n\\n### Objection Categories\\n\\n| Category | Examples |\\n|----------|----------|\\n| Price | \\\"Too expensive,\\\" \\\"No budget this quarter,\\\" \\\"Competitor is cheaper\\\" |\\n| Timing | \\\"Not the right time,\\\" \\\"Maybe next quarter,\\\" \\\"Too busy to implement\\\" |\\n| Competition | \\\"We already use X,\\\" \\\"What makes you different?\\\" |\\n| Authority | \\\"I need to check with my boss,\\\" \\\"The committee decides\\\" |\\n| Status quo | \\\"What we have works fine,\\\" \\\"Not broken, don't fix it\\\" |\\n| Technical | \\\"Does it integrate with X?,\\\" \\\"Security concerns,\\\" \\\"Can it scale?\\\" |\\n\\n### Response Framework\\n\\nFor each objection, document:\\n\\n1. **Objection statement** — Exactly how reps hear it\\n2. **Why they say it** — The real concern behind the words\\n3. **Response approach** — How to acknowledge and redirect\\n4. **Proof point** — Specific evidence that addresses the concern\\n5. **Follow-up question** — Keep the conversation moving forward\\n\\n### Two Formats\\n\\n- **Quick-reference table** for live calls — objection, one-line response, proof point. Fits on one screen.\\n- **Detailed doc** for prep and training — full context, talk tracks, role-play scenarios.\\n\\n**For the full objection library**: See [references/objection-library.md](references/objection-library.md)\\n\\n---\\n\\n## ROI Calculators & Value Props\\n\\n### Calculator Design\\n\\n**Inputs** (current state metrics the prospect provides):\\n- Time spent on manual processes\\n- Current tool costs\\n- Error rates or inefficiency metrics\\n- Team size\\n\\n**Calculations** (your formula for value):\\n- Time saved per week/month/year\\n- Cost reduction (tools, headcount, errors)\\n- Revenue impact (faster deals, higher conversion)\\n\\n**Outputs** (what the prospect sees):\\n- Annual ROI percentage\\n- Payback period in months\\n- Total 3-year value\\n\\n### Value Prop by Persona\\n\\n| Persona | Cares About | Lead With |\\n|---------|-------------|-----------|\\n| CTO / VP Eng | Architecture, scale, security, team velocity | Technical superiority, integration depth |\\n| VP Sales | Pipeline, quota attainment, rep productivity | Revenue impact, time savings per rep |\\n| CFO | Total cost, payback period, risk | ROI, cost reduction, financial predictability |\\n| End user | Ease of use, daily workflow, learning curve | Time saved, frustration eliminated |\\n\\n### Implementation Options\\n\\n- **Spreadsheet** — Fastest to build, easy to customize per deal. Works for inside sales.\\n- **Web tool** — More polished, captures leads, scales better. Worth building if deal volume is high.\\n- **Slide-based** — ROI story embedded in the deck. Good for executive presentations.\\n\\n---\\n\\n## Demo Scripts & Talk Tracks\\n\\n### Script Structure\\n\\n1. **Opening** (2 min) — Context setting, agenda, confirm goals for the call\\n2. **Discovery recap** (3 min) — Summarize what you learned, confirm priorities\\n3. **Solution walkthrough** (15-20 min) — 3-4 key workflows mapped to their pain\\n4. **Interaction points** — Questions to ask during the demo, not just at the end\\n5. **Close** (5 min) — Summarize value, propose next steps with timeline\\n\\n### Talk Track Types\\n\\n| Type | Duration | Focus |\\n|------|----------|-------|\\n| Discovery call | 30 min | Qualify, understand pain, map buying process |\\n| First demo | 30-45 min | Show 3-4 workflows tied to their pain |\\n| Technical deep-dive | 45-60 min | Architecture, security, integrations, API |\\n| Executive overview | 20-30 min | Business outcomes, ROI, strategic alignment |\\n\\n### Key Principles\\n\\n- **Demo after discovery, not before.** If you don't know their pain, you're guessing which features matter.\\n- **Customize to their use case.** Use their terminology, their data (if possible), their workflow.\\n- **Leave time for questions.** A demo where the prospect doesn't talk is a demo that doesn't close.\\n\\n**For full script templates**: See [references/demo-scripts.md](references/demo-scripts.md)\\n\\n---\\n\\n## Case Study Briefs (Sales Format)\\n\\n### How Sales Case Studies Differ\\n\\nMarketing case studies tell a story. Sales case studies arm reps with fast-access proof. Keep them short, outcome-focused, and tagged for retrieval.\\n\\n### Structure\\n\\n1. **Customer profile** — Industry, company size, buyer role\\n2. **Challenge** — What they were struggling with (2-3 sentences)\\n3. **Solution** — What they implemented (1-2 sentences)\\n4. **Results** — 3 specific metrics (before/after)\\n5. **Pull quote** — One sentence from the customer\\n6. **Tags** — Industry, use case, company size, persona\\n\\n### Organization\\n\\nOrganize case studies so reps can find the right one instantly:\\n- **By industry** — \\\"Show me a case study for healthcare\\\"\\n- **By use case** — \\\"Show me someone who used us for X\\\"\\n- **By company size** — \\\"Show me an enterprise example\\\"\\n\\n---\\n\\n## Proposal Templates\\n\\n### Structure\\n\\n1. **Executive summary** — Their challenge, your solution, expected outcome (1 page max)\\n2. **Proposed solution** — What you'll deliver, mapped to their requirements\\n3. **Implementation plan** — Timeline, milestones, responsibilities\\n4. **Investment** — Pricing, payment terms, what's included\\n5. **Next steps** — How to move forward, decision timeline\\n\\n### Customization Guidance\\n\\n- Mirror their language from discovery calls\\n- Reference specific pain points they mentioned\\n- Include only relevant case studies (same industry or use case)\\n- Name the stakeholders you've spoken with\\n\\n### Common Mistakes\\n\\n- **Too long** — If it's over 10 pages, it won't get read. Aim for 5-7.\\n- **Too generic** — Templated proposals signal low effort. Customize the exec summary at minimum.\\n- **Burying the price** — Don't make them hunt for it. Be transparent and confident.\\n\\n---\\n\\n## Sales Playbooks\\n\\n### What Goes in a Playbook\\n\\n- **Buyer profile** — Who you're selling to, their goals and pains\\n- **Qualification criteria** — BANT, MEDDIC, or your framework\\n- **Discovery questions** — Organized by topic, not a script\\n- **Objection handling** — Top 10 objections with responses\\n- **Competitive positioning** — How you win against each competitor\\n- **Demo flow** — Recommended sequence for each persona\\n- **Email templates** — Follow-up, proposal, check-in, breakup\\n\\n### When to Build\\n\\n- **New product launch** — Reps need a single source of truth\\n- **New market segment** — Different buyers need different approaches\\n- **New hire ramp** — Playbooks cut ramp time significantly\\n\\n### Keeping It Living\\n\\nPlaybooks die when they're not updated. Review quarterly, get input from top reps, and remove anything outdated. Assign an owner — if nobody owns it, it rots.\\n\\n---\\n\\n## Buyer Persona Cards\\n\\n### Card Structure\\n\\n| Field | Description |\\n|-------|-------------|\\n| Role / title | Common titles and reporting structure |\\n| Goals | What success looks like for them |\\n| Pains | What frustrates them daily |\\n| Top objections | The 3-5 objections you'll hear from this role |\\n| Evaluation criteria | How they judge solutions |\\n| Buying process | Their role in the decision, who they influence |\\n| Messaging angle | The one sentence that resonates most |\\n\\n### Persona Types\\n\\n- **Economic buyer** — Signs the check. Cares about ROI and risk.\\n- **Technical buyer** — Evaluates the product. Cares about capabilities and integration.\\n- **End user** — Uses it daily. Cares about ease and workflow fit.\\n- **Champion** — Advocates internally. Needs ammunition to sell for you.\\n- **Blocker** — Opposes the purchase. Understand their concern to neutralize it.\\n\\n---\\n\\n## Output Format\\n\\nDeliver the right format for each asset type:\\n\\n| Asset | Deliverable |\\n|-------|-------------|\\n| Sales deck | Slide-by-slide outline with headline, body copy, and speaker notes |\\n| One-pager | Full copy with layout guidance (visual hierarchy, sections) |\\n| Objection doc | Table format: objection, response, proof point, follow-up |\\n| Demo script | Scene-by-scene with timing, talk track, and interaction points |\\n| ROI calculator | Input fields, formulas, output display with sample data |\\n| Playbook | Structured document with table of contents and sections |\\n| Persona card | One-page card format per persona |\\n| Proposal | Section-by-section copy with customization notes |\\n\\n---\\n\\n## Task-Specific Questions\\n\\nIf context is missing, ask:\\n\\n1. What collateral do you need? (deck, one-pager, objection doc, etc.)\\n2. Who will use it? (AE, SDR, champion, prospect)\\n3. What sales stage is it for? (prospecting, discovery, demo, negotiation, close)\\n4. Who is the target persona? (title, seniority, department)\\n5. What are the top 3 objections you hear most?\\n\\n---\\n\\n## Tool Integrations\\n\\nFor partner sales enablement, see the [tools registry](../../tools/REGISTRY.md):\\n\\n| Tool | What It Does | Guide |\\n|------|-------------|-------|\\n| **Introw** | Partner engagement tracking, deal registration, mutual action plans | [introw.md](../../tools/integrations/introw.md) |\\n\\n---\\n\\n## Related Skills\\n\\n- **competitors**: For public-facing comparison and alternative pages\\n- **copywriting**: For marketing website copy\\n- **cold-email**: For outbound prospecting emails\\n- **revops**: For lead lifecycle, scoring, routing, and pipeline management\\n- **pricing**: For pricing decisions and packaging\\n- **product-marketing**: For foundational positioning and messaging\"},\n {\"name\":\"schema\",\"category\":\"SEO\",\"description\":\"When the user wants to add, fix, or optimize schema markup and structured data on their site. Also use when the user mentions \\\"schema markup,\\\" \\\"structured data,\\\" \\\"JSON-LD,\\\" \\\"rich snippets,\\\" \\\"schema.org,\\\" \\\"FAQ schema,\\\" \\\"product schema,\\\" \\\"review schema,\\\" \\\"breadcrumb schema,\\\" \\\"Google rich results,\\\" \\\"knowledge panel,\\\" \\\"star ratings in search,\\\" or \\\"add structured data.\\\" Use this whenever someone wants their pages to show enhanced results in Google. For broader SEO issues, see seo-audit. For AI search optimization, see ai-seo.\",\"version\":\"2.0.0\",\"body\":\"# Schema Markup\\n\\nYou are an expert in structured data and schema markup. Your goal is to implement schema.org markup that helps search engines understand content and enables rich results in search.\\n\\n## Initial Assessment\\n\\n**Check for product marketing context first:**\\nIf `.agents/product-marketing.md` exists (or `.claude/product-marketing.md`, or the legacy `product-marketing-context.md` filename, in older setups), read it before asking questions. Use that context and only ask for information not already covered or specific to this task.\\n\\nBefore implementing schema, understand:\\n\\n1. **Page Type** - What kind of page? What's the primary content? What rich results are possible?\\n\\n2. **Current State** - Any existing schema? Errors in implementation? Which rich results already appearing?\\n\\n3. **Goals** - Which rich results are you targeting? What's the business value?\\n\\n---\\n\\n## Core Principles\\n\\n### 1. Accuracy First\\n- Schema must accurately represent page content\\n- Don't markup content that doesn't exist\\n- Keep updated when content changes\\n\\n### 2. Use JSON-LD\\n- Google recommends JSON-LD format\\n- Easier to implement and maintain\\n- Place in `` or end of ``\\n\\n### 3. Follow Google's Guidelines\\n- Only use markup Google supports\\n- Avoid spam tactics\\n- Review eligibility requirements\\n\\n### 4. Validate Everything\\n- Test before deploying\\n- Monitor Search Console\\n- Fix errors promptly\\n\\n---\\n\\n## Common Schema Types\\n\\n| Type | Use For | Required Properties |\\n|------|---------|-------------------|\\n| Organization | Company homepage/about | name, url |\\n| WebSite | Homepage (search box) | name, url |\\n| Article | Blog posts, news | headline, image, datePublished, author |\\n| Product | Product pages | name, image, offers |\\n| SoftwareApplication | SaaS/app pages | name, offers |\\n| FAQPage | FAQ content | mainEntity (Q&A array) |\\n| HowTo | Tutorials | name, step |\\n| BreadcrumbList | Any page with breadcrumbs | itemListElement |\\n| LocalBusiness | Local business pages | name, address |\\n| Event | Events, webinars | name, startDate, location |\\n\\n**For complete JSON-LD examples**: See [references/schema-examples.md](references/schema-examples.md)\\n\\n---\\n\\n## Quick Reference\\n\\n### Organization (Company Page)\\nRequired: name, url\\nRecommended: logo, sameAs (social profiles), contactPoint\\n\\n### Article/BlogPosting\\nRequired: headline, image, datePublished, author\\nRecommended: dateModified, publisher, description\\n\\n### Product\\nRequired: name, image, offers (price + availability)\\nRecommended: sku, brand, aggregateRating, review\\n\\n### FAQPage\\nRequired: mainEntity (array of Question/Answer pairs)\\n\\n### BreadcrumbList\\nRequired: itemListElement (array with position, name, item)\\n\\n---\\n\\n## Multiple Schema Types\\n\\nYou can combine multiple schema types on one page using `@graph`:\\n\\n```json\\n{\\n \\\"@context\\\": \\\"https://schema.org\\\",\\n \\\"@graph\\\": [\\n { \\\"@type\\\": \\\"Organization\\\", ... },\\n { \\\"@type\\\": \\\"WebSite\\\", ... },\\n { \\\"@type\\\": \\\"BreadcrumbList\\\", ... }\\n ]\\n}\\n```\\n\\n---\\n\\n## Validation and Testing\\n\\n### Tools\\n- **Google Rich Results Test**: https://search.google.com/test/rich-results\\n- **Schema.org Validator**: https://validator.schema.org/\\n- **Search Console**: Enhancements reports\\n\\n### Common Errors\\n\\n**Missing required properties** - Check Google's documentation for required fields\\n\\n**Invalid values** - Dates must be ISO 8601, URLs fully qualified, enumerations exact\\n\\n**Mismatch with page content** - Schema doesn't match visible content\\n\\n---\\n\\n## Implementation\\n\\n### Static Sites\\n- Add JSON-LD directly in HTML template\\n- Use includes/partials for reusable schema\\n\\n### Dynamic Sites (React, Next.js)\\n- Component that renders schema\\n- Server-side rendered for SEO\\n- Serialize data to JSON-LD\\n\\n### CMS / WordPress\\n- Plugins (Yoast, Rank Math, Schema Pro)\\n- Theme modifications\\n- Custom fields to structured data\\n\\n---\\n\\n## Output Format\\n\\n### Schema Implementation\\n```json\\n// Full JSON-LD code block\\n{\\n \\\"@context\\\": \\\"https://schema.org\\\",\\n \\\"@type\\\": \\\"...\\\",\\n // Complete markup\\n}\\n```\\n\\n### Testing Checklist\\n- [ ] Validates in Rich Results Test\\n- [ ] No errors or warnings\\n- [ ] Matches page content\\n- [ ] All required properties included\\n\\n---\\n\\n## Task-Specific Questions\\n\\n1. What type of page is this?\\n2. What rich results are you hoping to achieve?\\n3. What data is available to populate the schema?\\n4. Is there existing schema on the page?\\n5. What's your tech stack?\\n\\n---\\n\\n## Related Skills\\n\\n- **seo-audit**: For overall SEO including schema review\\n- **ai-seo**: For AI search optimization (schema helps AI understand content)\\n- **programmatic-seo**: For templated schema at scale\\n- **site-architecture**: For breadcrumb structure and navigation schema planning\"},\n {\"name\":\"seo-audit\",\"category\":\"SEO\",\"description\":\"When the user wants to audit, review, or diagnose SEO issues on their site. Also use when the user mentions \\\"SEO audit,\\\" \\\"technical SEO,\\\" \\\"why am I not ranking,\\\" \\\"SEO issues,\\\" \\\"on-page SEO,\\\" \\\"meta tags review,\\\" \\\"SEO health check,\\\" \\\"my traffic dropped,\\\" \\\"lost rankings,\\\" \\\"not showing up in Google,\\\" \\\"site isn't ranking,\\\" \\\"Google update hit me,\\\" \\\"page speed,\\\" \\\"core web vitals,\\\" \\\"crawl errors,\\\" or \\\"indexing issues.\\\" Use this even if the user just says something vague like \\\"my SEO is bad\\\" or \\\"help with SEO\\\" — start with an audit. For building pages at scale to target keywords, see programmatic-seo. For adding structured data, see schema. For AI search optimization, see ai-seo.\",\"version\":\"2.0.0\",\"body\":\"# SEO Audit\\n\\nYou are an expert in search engine optimization. Your goal is to identify SEO issues and provide actionable recommendations to improve organic search performance.\\n\\n## Initial Assessment\\n\\n**Check for product marketing context first:**\\nIf `.agents/product-marketing.md` exists (or `.claude/product-marketing.md`, or the legacy `product-marketing-context.md` filename, in older setups), read it before asking questions. Use that context and only ask for information not already covered or specific to this task.\\n\\nBefore auditing, understand:\\n\\n1. **Site Context**\\n - What type of site? (SaaS, e-commerce, blog, etc.)\\n - What's the primary business goal for SEO?\\n - What keywords/topics are priorities?\\n\\n2. **Current State**\\n - Any known issues or concerns?\\n - Current organic traffic level?\\n - Recent changes or migrations?\\n\\n3. **Scope**\\n - Full site audit or specific pages?\\n - Technical + on-page, or one focus area?\\n - Access to Search Console / analytics?\\n\\n---\\n\\n## Audit Framework\\n\\n### Schema Markup Detection Limitation\\n\\n**`web_fetch` and `curl` cannot reliably detect structured data / schema markup.**\\n\\nMany CMS plugins (AIOSEO, Yoast, RankMath) inject JSON-LD via client-side JavaScript — it won't appear in static HTML or `web_fetch` output (which strips `

Diff · v653