Taskflow AI | Intelligent Task Automation

[{"_1":2,"_3":-5,"_4":-5},"loaderData",{"_5":6,"_7":8,"_9":10},"actionData","errors","root",{"_153":154,"_155":156},"routes/_app.x.$handle.$name",{"_95":94,"_32":45,"_103":85,"_27":104,"_13":14,"_105":-5,"_11":106,"_107":108,"_109":110,"_111":112,"_113":114,"_20":-5,"_115":-5,"_116":-5,"_117":25,"_118":119,"_120":57,"_121":-5,"_122":-5,"_123":124},"routes/_app.x.$handle.$name.branch.$branch.version.$version.diff",{"_11":12,"_13":14,"_15":16,"_17":-5,"_18":-5},"project",{"_43":44,"_32":45,"_38":46,"_47":-5,"_48":25,"_49":50,"_51":50,"_52":-5,"_53":-5,"_18":54,"_55":31,"_56":57,"_58":57,"_59":-5,"_60":61,"_62":63,"_64":-5,"_65":-5,"_66":-5,"_67":-5,"_68":69,"_70":69,"_71":72,"_73":72,"_74":72,"_75":69,"_76":69,"_77":78,"_79":80,"_81":82,"_83":25},"version",88,"changes",[19],"aiChatInfo","userId",{"_20":21,"_22":23,"_24":25,"_26":25,"_27":28,"_29":28},"file",{"_30":31,"_32":28,"_33":34,"_35":37,"_38":39,"_40":14,"_41":42,"_24":25},"previous",{"_30":31,"_32":28,"_33":34,"_35":36},"deleted",false,"deletedHere","path","main.ts","latestPath","fileId","4b285790-a760-11f0-b29b-0224a6c84d84","name","fileType","http","content","// @ts-nocheck\nimport { Hono } from \"npm:hono@4.4.12\";\nimport { OpenAI } from \"https://esm.town/v/std/openai\";\nimport type { Context } from \"npm:hono@4.4.12\";\nimport { streamText } from \"npm:hono@4.4.12/streaming\";\n\ninterface CompanyContext {\n company_name: string;\n company_description: string;\n use_case: string;\n language: string;\n industry?: string;\n}\ninterface ConversationHistory {\n role: \"assistant\" | \"user\";\n content: string;\n}\ninterface GatherBody {\n task: string;\n domain: string;\n history: ConversationHistory[];\n}\ninterface DynamicPromptBody {\n occupation_title: string;\n task: string;\n company_context: CompanyContext;\n}\ninterface ExecuteBody {\n refined_prompt: string;\n user_inputs: Record;\n company_context: CompanyContext;\n category: string;\n}\n\nconst config = {\n models: {\n major: \"gpt-4o\",\n minor: \"gpt-4o-mini\",\n },\n prompts: {\n INFORMATION_GATHERER:\n `You are a native JSON speaking expert information gatherer designed to acquire all necessary details for a given task and domain. Your goal is to ask the fewest, most concise, and natural questions possible, one at a time. If information is missing or refused, make a reasonable assumption or broaden your scope. Iteratively ask questions and integrate responses until all required information for the {{task}} within the {{domain}} is gathered.\nRULES\nPrimary Goal: Ask single, concise questions to gather the minimum required information for the task, aiming for completion efficiently.\nDecision Point: After analyzing the conversation history, decide if you have enough information.If YES: Your immediate response MUST be the token \"COMPLETE\" followed by a single, minified JSON object of all gathered data. Do not ask for confirmation. Example: ` +\n \"`COMPLETE`\" +\n `{\"party_date\":\"next Saturday\",\"guests\":50,\"location\":\"rented hall\"}\nIf NO: You are missing critical information. Proceed to the next rule.\nQuestion Formulation: If more info is needed, formulate the single most impactful and concise question.\nOutput Format (Questions): The question must be a short, direct interrogative. DO NOT add conversational filler, commentary, apologies, or instructions.\nCONTEXT\nTask: {{task}}\nDomain: {{domain}}\nConversation History:\n{{history}}`,\n PROMPT_REFINER:\n `You are a native JSON speaking prompt engineer. Transmute raw occupational data into a concise, forceful directive for a target AI. The AI must perform the task end-to-end and return only the finished deliverable(s)—never a plan. You will receive an \"Occupation,\" \"Task,\" and \"Company Context.\" Your prompt must be a production-grade instruction that directly yields the final artifact. CORE RULE: You MUST return a prompt with {{variables}}. Your goal is to identify important, undefined entities in the Task and convert them to descriptive, {{snake_case_placeholders}}. You MUST include at least 2 placeholders. INSTRUCTIONS: 1. Persona: Start with \"You are a top 0.1% [Occupation]...\" for an expert tone. 2. Objective: Translate the Task into a formal, actionable objective implying completion. 3. Context: Use the 'Company Context' to make the objective specific (e.g., \"Your mandate for [Company Name] is to...\"). 4. Placeholders: Convert only USER-SUPPLIED functional inputs into {{snake_case_variables}}. 5. Delivery Rules: Require the AI to: - Return only the finished deliverable(s) now. - Exclude plans, methods, or explanations. - Avoid meta-commentary, caveats, or apologies. - If details are missing, make minimal, reasonable assumptions and proceed, optionally adding a short \"Assumptions\" list (<=5 bullets). - Use production-ready formatting (e.g., ready-to-send copy; runnable code in fenced blocks). - Never include sections like \"Plan\", \"Approach\", \"Methodology\". 6. Output Format: Specify the exact output structure. 7. Mandate: Conclude with a clear call to action like \"Produce the deliverable now.\" Your entire response MUST be a single, minified JSON object with one key, \"refined_prompt\". No other text or markdown.`,\n MASTER_PROMPT_ENGINEER:\n `You are a native JSON speaking master prompt architect. Your objective is to construct a comprehensive, self-contained blueprint for another AI to execute a task flawlessly, not to do it yourself.\nINSTRUCTIONS\nDeconstruct Task: Analyze the task to identify sub-tasks, knowledge domains, constraints, and desired output formats.\nEstablish Persona: Define an expert persona for the target AI (e.g., \"You are a tenured professor of economics...\").\nProvide Context: Synthesize company context into a rich narrative explaining the task's strategic importance.\nFormulate Logic: Create a clear, logical sequence of operations using headings and lists.\nSet Negative Constraints: Explicitly state what the AI should not do (e.g., \"Do not use informal language,\" \"Do not output a plan\").\nDefine Output Schema: Meticulously define the final output's structure, format, length, and tone.\nInclude Examples: Where helpful, provide concise examples of good and bad outputs.\nFinal Mandate: Conclude with a clear command to initiate execution.\nYour output must be the prompt itself, as a clean markdown document. Do not wrap it in JSON or add commentary.`,\n INDUSTRY_GENERATOR:\n `You are a native JSON speaking expert business analyst. Generate a list of professional industries most relevant to the given company.\nRULES\nAnalyze Context: Deeply analyze the provided company context (name, description, use case) to understand their core business.\nGenerate Industries: Produce a list of 10-12 diverse yet highly relevant industries.\nOutput Format: Your entire response MUST be a single, minified JSON object with one key: \"industries\". The value must be an array of strings.\nCOMPANY CONTEXT\n{{company_context}}`,\n DYNAMIC_LIST_GENERATOR:\n `You are a native JSON speaking expert AI that generates context-aware lists for professional applications. You will receive a JSON object with company context and the list type to generate.\nRULES\nAnalyze Context: Deeply analyze the 'company_context' to understand their business.\nGenerate List: Create a list of 10-15 relevant items based on 'generation_type'.If 'generation_type' is \"occupations\", generate specific job titles within the 'industry'.\nIf 'generation_type' is \"tasks\", generate specific, actionable tasks for the 'occupation'.\nOutput Format: Your response MUST be a single, minified JSON object with one key: \"generated_list\".\nList Structure: The value must be an array of objects, each with \"name\" (string) and \"description\" (concise string).\nRelevance is Key: Items must be sharply focused and directly applicable to the company's use case.\nInput is JSON: You must parse the JSON input to get the context.`,\n EVALUATOR_AGENT:\n `You are a native JSON speaking, meticulous, and impartial AI QA agent. Given a task briefing, success criteria, and AI output, you will:\nScore: Score the output 1-10 on each criterion.\nCritique: Provide a concise, professional critique of its strengths and weaknesses.\nSCORING RULES\nGut Check: Before all else, if the output has obvious placeholders ('example.com', '[Insert Name Here]') or is extremely generic, penalize heavily on \"Plausibility and Specificity.\"\nPenalize Non-Deliverables: If the output is a plan or commentary instead of the final deliverable, penalize heavily (scores <= 3) on the relevant criterion.\nOUTPUT FORMAT\nYour entire response MUST be a single, minified JSON object: {\"scores\": [{\"criterion\": \"...\", \"score\": ...}], \"critique\": \"...\"}. You MUST write the critique in the requested language. Add no other text or markdown.`,\n REFINER_AGENT:\n `You are a top-tier \"Refiner\" agent. You have your original task briefing, your first draft, and a critique from a QA agent. Your SOLE mission is to rewrite the first draft to fully address every point in the critique, elevating it to a 10/10 standard.\nRULES\nAnalyze and Integrate: Apply all suggested changes from the critique to the first draft.\nNo Commentary: Do not apologize, explain changes, or include meta commentary.\nFinal Deliverable Only: Your output must be only the new, improved, final version.\nLanguage: Write the final output in the requested language.`,\n SUCCESS_CRITERIA_GENERATOR:\n `You are a native JSON speaking expert QA analyst. Your task is to generate a concise list of success criteria for evaluating an AI-generated document.\nRULES\nAnalyze Task: Deeply analyze the task briefing to understand the desired output's purpose and requirements.\nGenerate Criteria: Create 4-6 specific, measurable, relevant success criteria as short strings.\nOutput Format: Your response MUST be a single, minified JSON object: {\"criteria\": [\"...\", \"...\", ...]}.\nTASK BRIEFING\n{{task_briefing}}`,\n },\n};\n\nconst taskCategories = [\n {\n id: \"thinking\",\n name: \"Strategic Planning\",\n description: \"Generate strategies, outlines, or comprehensive analyses.\",\n icon: \"🧠\",\n },\n {\n id: \"prompt_building\",\n name: \"Prompt Engineering\",\n description: \"Create detailed, expert-level prompts for AI systems.\",\n icon: \"🛠️\",\n },\n {\n id: \"document_creation\",\n name: \"Professional Documents\",\n description:\n \"Produce polished, ready-to-use documents with quality refinement.\",\n icon: \"📄\",\n },\n];\n\nconst services = {\n async callOpenAI(\n systemPrompt: string,\n userContent: string | object,\n options: {\n model?: string;\n isJson?: boolean;\n stream?: boolean;\n c?: Context;\n } = {},\n ) {\n const {\n model = config.models.major,\n isJson = false,\n stream = false,\n c = null,\n } = options;\n const openai = new OpenAI();\n const messages: any[] = [{ role: \"system\", content: systemPrompt }, {\n role: \"user\",\n content: typeof userContent === \"string\"\n ? userContent\n : JSON.stringify(userContent),\n }];\n const requestPayload: any = { model, messages, stream };\n if (isJson) requestPayload.response_format = { type: \"json_object\" };\n try {\n const completion = await openai.chat.completions.create(requestPayload);\n if (stream && c) {\n return streamText(c, async (s) => {\n for await (const part of completion) {\n await s.write(part.choices[0]?.delta?.content || \"\");\n }\n });\n }\n if (isJson && c) {\n return c.json(JSON.parse(completion.choices[0].message.content));\n }\n return completion;\n } catch (e) {\n console.error(`Error calling OpenAI: ${e.message}`);\n throw new Error(\"AI service failed.\");\n }\n },\n};\n\nasync function handleExecute(c: Context) {\n try {\n const body: ExecuteBody = await c.req.json();\n const { refined_prompt, user_inputs, company_context, category } = body;\n\n let final_user_prompt = refined_prompt;\n for (const key in user_inputs) {\n const regex = new RegExp(`{{${key}}}`, \"g\");\n final_user_prompt = final_user_prompt.replace(regex, user_inputs[key]);\n }\n\n switch (category) {\n case \"thinking\":\n return services.callOpenAI(\n final_user_prompt,\n \"Execute the task based on the system prompt. Proceed.\",\n { model: config.models.major, stream: true, c },\n );\n\n case \"prompt_building\":\n return services.callOpenAI(\n config.prompts.MASTER_PROMPT_ENGINEER,\n final_user_prompt,\n { model: config.models.major, stream: true, c },\n );\n\n case \"document_creation\":\n const draftCompletion = await services.callOpenAI(\n final_user_prompt,\n \"Execute the task based on the system prompt. Return only the deliverable.\",\n { model: config.models.major },\n );\n const first_draft = draftCompletion.choices[0].message.content;\n\n const criteriaSystemPrompt = config.prompts.SUCCESS_CRITERIA_GENERATOR\n .replace(\"{{task_briefing}}\", final_user_prompt);\n\n const criteriaCompletion = await services.callOpenAI(\n criteriaSystemPrompt,\n \"Generate.\",\n { model: config.models.minor, isJson: true },\n );\n const criteria_json = JSON.parse(\n criteriaCompletion.choices[0].message.content,\n );\n const criteria = criteria_json.criteria;\n\n const evalUserContent = {\n task_briefing: final_user_prompt,\n success_criteria: criteria,\n ai_output: first_draft,\n language: company_context.language,\n };\n\n const evalCompletion = await services.callOpenAI(\n config.prompts.EVALUATOR_AGENT,\n evalUserContent,\n { model: config.models.minor, isJson: true },\n );\n const critique_json = JSON.parse(\n evalCompletion.choices[0].message.content,\n );\n const critique = critique_json.critique;\n\n const refineUserContent = {\n task_briefing: final_user_prompt,\n first_draft: first_draft,\n critique: critique,\n language: company_context.language,\n };\n\n return services.callOpenAI(\n config.prompts.REFINER_AGENT,\n refineUserContent,\n { model: config.models.major, stream: true, c },\n );\n\n default:\n return c.json({ error: \"Invalid task category\" }, 400);\n }\n } catch (e) {\n console.error(`Error in handleExecute: ${e.message}`);\n return c.json({ error: \"Task execution failed.\", details: e.message }, 500);\n }\n}\n\nfunction generateHtml(sourceUrl: string): string {\n const languages = {\n \"English\": \"en-US\",\n \"Spanish\": \"es-ES\",\n \"French\": \"fr-FR\",\n \"German\": \"de-DE\",\n \"Japanese\": \"ja-JP\",\n \"Chinese (Simplified)\": \"zh-CN\",\n };\n const languageOptions = Object.entries(languages).map(([name, code]) =>\n ``\n ).join(\"\");\n const taskCategoriesJson = JSON.stringify(taskCategories);\n return `Taskflow AI | Intelligent Task Automation\n \n \n \n \n \n \n \n \n \n

\n Great job!\n

\n You're making excellent progress.\n

\n \n \n \n

\n \n \n \n \n

\n You're doing amazing! ✨\n

\n Taskflow AI is free and open-source. If you find it valuable,\n consider supporting its development. Your generosity helps\n keep this tool free for everyone!\n

\n ☕ Buy me a coffee

\n This won't show again for 7 days\n

\n \n \n \n

\n \n Welcome to Taskflow AI\n \n

\n Let's get you set up in just a few moments\n

\n \n

\n What makes Taskflow AI special?\n

\n \n AI agents tailored to your industry and needs\n
\n \n Automatic quality refinement for professional\n results\n
\n \n Conversational interface - just answer a few\n questions\n

\n \n

Quick Setup

\n Step 1 of 2\n

\n \n \n

\n Almost there!\n

\n Step 2 of 2\n

\n What best describes your primary use case?\n

\n \n

\n 📣 Marketing & Content\n

\n Create campaigns, copy, and content\n

\n ⚙️ Operations & Strategy\n

\n Plans, processes, and documentation\n

\n 💻 Product & Development\n

\n Technical documentation, specs, and prompts\n

\n ✨ General Business Tasks\n

\n Various professional documents and analyses\n

\n \n

\n \n \n \n \n \n \n \n

\n You're all set! 🎉\n

\n Your workspace is ready. Let's create something amazing\n together!\n

\n \n \n \n \n

\n \n \n \n \n \n \n \n \n TASKFLOW\n \n \n \n \n \n \n \n

\n \n Your Stats\n

\n \n \n 0\n

Tasks

\n \n \n \n 0\n \n

\n Refined\n

\n \n \n \n

\n \n Support Taskflow\n \n

\n \n

\n \n \n \n Workspace Settings\n \n

\n \n

\n \n \n \n Taskflow AI\n \n \n \n

\n \n \n \n \n \n`;\n }\n\nconst app = new Hono();\n\napp.get(\"/\", (c) => c.html(generateHtml(c.req.url)));\n\napp.post(\"/api/generate-industries\", async (c) => {\n try {\n const { company_context } = await c.req.json();\n const systemPrompt = config.prompts.INDUSTRY_GENERATOR.replace(\n \"{{company_context}}\",\n JSON.stringify(company_context),\n );\n return await services.callOpenAI(systemPrompt, \"Generate.\", {\n model: config.models.minor,\n isJson: true,\n c: c,\n });\n } catch (e) {\n return c.json({ error: e.message }, 500);\n }\n});\n\napp.post(\"/api/generate-dynamic-list\", async (c) => {\n try {\n const body = await c.req.json();\n return await services.callOpenAI(\n config.prompts.DYNAMIC_LIST_GENERATOR,\n body,\n { model: config.models.minor, isJson: true, c: c },\n );\n } catch (e) {\n return c.json({ error: e.message }, 500);\n }\n});\n\napp.post(\"/api/gather\", async (c) => {\n try {\n const { task, domain, history }: GatherBody = await c.req.json();\n const historyStr = history.map((h) => `${h.role}: ${h.content}`).join(\"\\n\");\n let systemPrompt = config.prompts.INFORMATION_GATHERER\n .replace(\"{{task}}\", task)\n .replace(\"{{domain}}\", domain)\n .replace(\"{{history}}\", historyStr);\n\n const completion = await services.callOpenAI(\n systemPrompt,\n \"Continue the information gathering process.\",\n { model: config.models.minor },\n );\n\n let content = completion.choices[0].message.content;\n\n if (content.startsWith(\"COMPLETE\")) {\n const jsonData = content.substring(8).trim();\n return c.json({ complete: true, data: JSON.parse(jsonData) });\n }\n\n return streamText(c, async (s) => {\n await s.write(content);\n });\n } catch (e) {\n return c.json({ error: e.message }, 500);\n }\n});\n\napp.post(\"/api/refine-prompt\", async (c) => {\n try {\n const body: DynamicPromptBody = await c.req.json();\n const { occupation_title, task, company_context } = body;\n const userContent = {\n occupation_title: occupation_title,\n task: task,\n company_context: company_context,\n };\n return await services.callOpenAI(\n config.prompts.PROMPT_REFINER,\n userContent,\n { model: config.models.major, isJson: true, c: c },\n );\n } catch (e) {\n return c.json({ error: e.message }, 500);\n }\n});\n\napp.post(\"/api/execute\", handleExecute);\n\nexport default app.fetch;","// @ts-nocheck\nimport { Hono } from \"npm:hono@4.4.12\";\nimport { OpenAI } from \"https://esm.town/v/std/openai\";\nimport type { Context } from \"npm:hono@4.4.12\";\nimport { streamText } from \"npm:hono@4.4.12/streaming\";\n\ninterface CompanyContext {\n company_name: string;\n company_description: string;\n use_case: string;\n language: string;\n industry?: string;\n}\ninterface ConversationHistory {\n role: \"assistant\" | \"user\";\n content: string;\n}\ninterface GatherBody {\n task: string;\n domain: string;\n history: ConversationHistory[];\n}\ninterface DynamicPromptBody {\n occupation_title: string;\n task: string;\n company_context: CompanyContext;\n}\ninterface ExecuteBody {\n refined_prompt: string;\n user_inputs: Record;\n company_context: CompanyContext;\n category: string;\n}\n\nconst config = {\n models: {\n major: \"gpt-4o\",\n minor: \"gpt-4o-mini\",\n },\n prompts: {\n INFORMATION_GATHERER:\n `You are a native JSON speaking expert information gatherer designed to acquire all necessary details for a given task and domain. Your goal is to ask the fewest, most concise, and natural questions possible, one at a time. If information is missing or refused, make a reasonable assumption or broaden your scope. Iteratively ask questions and integrate responses until all required information for the {{task}} within the {{domain}} is gathered.\nRULES\nPrimary Goal: Ask single, concise questions to gather the minimum required information for the task, aiming for completion efficiently.\nDecision Point: After analyzing the conversation history, decide if you have enough information.If YES: Your immediate response MUST be the token \"COMPLETE\" followed by a single, minified JSON object of all gathered data. Do not ask for confirmation. Example: ` +\n \"`COMPLETE`\" +\n `{\"party_date\":\"next Saturday\",\"guests\":50,\"location\":\"rented hall\"}\nIf NO: You are missing critical information. Proceed to the next rule.\nQuestion Formulation: If more info is needed, formulate the single most impactful and concise question.\nOutput Format (Questions): The question must be a short, direct interrogative. DO NOT add conversational filler, commentary, apologies, or instructions.\nCONTEXT\nTask: {{task}}\nDomain: {{domain}}\nConversation History:\n{{history}}`,\n PROMPT_REFINER:\n `You are a native JSON speaking prompt engineer. Transmute raw occupational data into a concise, forceful directive for a target AI. The AI must perform the task end-to-end and return only the finished deliverable(s)—never a plan. You will receive an \"Occupation,\" \"Task,\" and \"Company Context.\" Your prompt must be a production-grade instruction that directly yields the final artifact. CORE RULE: You MUST return a prompt with {{variables}}. Your goal is to identify important, undefined entities in the Task and convert them to descriptive, {{snake_case_placeholders}}. You MUST include at least 2 placeholders. INSTRUCTIONS: 1. Persona: Start with \"You are a top 0.1% [Occupation]...\" for an expert tone. 2. Objective: Translate the Task into a formal, actionable objective implying completion. 3. Context: Use the 'Company Context' to make the objective specific (e.g., \"Your mandate for [Company Name] is to...\"). 4. Placeholders: Convert only USER-SUPPLIED functional inputs into {{snake_case_variables}}. 5. Delivery Rules: Require the AI to: - Return only the finished deliverable(s) now. - Exclude plans, methods, or explanations. - Avoid meta-commentary, caveats, or apologies. - If details are missing, make minimal, reasonable assumptions and proceed, optionally adding a short \"Assumptions\" list (<=5 bullets). - Use production-ready formatting (e.g., ready-to-send copy; runnable code in fenced blocks). - Never include sections like \"Plan\", \"Approach\", \"Methodology\". 6. Output Format: Specify the exact output structure. 7. Mandate: Conclude with a clear call to action like \"Produce the deliverable now.\" Your entire response MUST be a single, minified JSON object with one key, \"refined_prompt\". No other text or markdown.`,\n MASTER_PROMPT_ENGINEER:\n `You are a native JSON speaking master prompt architect. Your objective is to construct a comprehensive, self-contained blueprint for another AI to execute a task flawlessly, not to do it yourself.\nINSTRUCTIONS\nDeconstruct Task: Analyze the task to identify sub-tasks, knowledge domains, constraints, and desired output formats.\nEstablish Persona: Define an expert persona for the target AI (e.g., \"You are a tenured professor of economics...\").\nProvide Context: Synthesize company context into a rich narrative explaining the task's strategic importance.\nFormulate Logic: Create a clear, logical sequence of operations using headings and lists.\nSet Negative Constraints: Explicitly state what the AI should not do (e.g., \"Do not use informal language,\" \"Do not output a plan\").\nDefine Output Schema: Meticulously define the final output's structure, format, length, and tone.\nInclude Examples: Where helpful, provide concise examples of good and bad outputs.\nFinal Mandate: Conclude with a clear command to initiate execution.\nYour output must be the prompt itself, as a clean markdown document. Do not wrap it in JSON or add commentary.`,\n INDUSTRY_GENERATOR:\n `You are a native JSON speaking expert business analyst. Generate a list of professional industries most relevant to the given company.\nRULES\nAnalyze Context: Deeply analyze the provided company context (name, description, use case) to understand their core business.\nGenerate Industries: Produce a list of 10-12 diverse yet highly relevant industries.\nOutput Format: Your entire response MUST be a single, minified JSON object with one key: \"industries\". The value must be an array of strings.\nCOMPANY CONTEXT\n{{company_context}}`,\n DYNAMIC_LIST_GENERATOR:\n `You are a native JSON speaking expert AI that generates context-aware lists for professional applications. You will receive a JSON object with company context and the list type to generate.\nRULES\nAnalyze Context: Deeply analyze the 'company_context' to understand their business.\nGenerate List: Create a list of 10-15 relevant items based on 'generation_type'.If 'generation_type' is \"occupations\", generate specific job titles within the 'industry'.\nIf 'generation_type' is \"tasks\", generate specific, actionable tasks for the 'occupation'.\nOutput Format: Your response MUST be a single, minified JSON object with one key: \"generated_list\".\nList Structure: The value must be an array of objects, each with \"name\" (string) and \"description\" (concise string).\nRelevance is Key: Items must be sharply focused and directly applicable to the company's use case.\nInput is JSON: You must parse the JSON input to get the context.`,\n EVALUATOR_AGENT:\n `You are a native JSON speaking, meticulous, and impartial AI QA agent. Given a task briefing, success criteria, and AI output, you will:\nScore: Score the output 1-10 on each criterion.\nCritique: Provide a concise, professional critique of its strengths and weaknesses.\nSCORING RULES\nGut Check: Before all else, if the output has obvious placeholders ('example.com', '[Insert Name Here]') or is extremely generic, penalize heavily on \"Plausibility and Specificity.\"\nPenalize Non-Deliverables: If the output is a plan or commentary instead of the final deliverable, penalize heavily (scores <= 3) on the relevant criterion.\nOUTPUT FORMAT\nYour entire response MUST be a single, minified JSON object: {\"scores\": [{\"criterion\": \"...\", \"score\": ...}], \"critique\": \"...\"}. You MUST write the critique in the requested language. Add no other text or markdown.`,\n REFINER_AGENT:\n `You are a top-tier \"Refiner\" agent. You have your original task briefing, your first draft, and a critique from a QA agent. Your SOLE mission is to rewrite the first draft to fully address every point in the critique, elevating it to a 10/10 standard.\nRULES\nAnalyze and Integrate: Apply all suggested changes from the critique to the first draft.\nNo Commentary: Do not apologize, explain changes, or include meta commentary.\nFinal Deliverable Only: Your output must be only the new, improved, final version.\nLanguage: Write the final output in the requested language.`,\n SUCCESS_CRITERIA_GENERATOR:\n `You are a native JSON speaking expert QA analyst. Your task is to generate a concise list of success criteria for evaluating an AI-generated document.\nRULES\nAnalyze Task: Deeply analyze the task briefing to understand the desired output's purpose and requirements.\nGenerate Criteria: Create 4-6 specific, measurable, relevant success criteria as short strings.\nOutput Format: Your response MUST be a single, minified JSON object: {\"criteria\": [\"...\", \"...\", ...]}.\nTASK BRIEFING\n{{task_briefing}}`,\n },\n};\n\nconst taskCategories = [\n {\n id: \"thinking\",\n name: \"Strategic Planning\",\n description: \"Generate strategies, outlines, or comprehensive analyses.\",\n icon: \"🧠\",\n },\n {\n id: \"prompt_building\",\n name: \"Prompt Engineering\",\n description: \"Create detailed, expert-level prompts for AI systems.\",\n icon: \"🛠️\",\n },\n {\n id: \"document_creation\",\n name: \"Professional Documents\",\n description:\n \"Produce polished, ready-to-use documents with quality refinement.\",\n icon: \"📄\",\n },\n];\n\nconst services = {\n async callOpenAI(\n systemPrompt: string,\n userContent: string | object,\n options: {\n model?: string;\n isJson?: boolean;\n stream?: boolean;\n c?: Context;\n } = {},\n ) {\n const {\n model = config.models.major,\n isJson = false,\n stream = false,\n c = null,\n } = options;\n const openai = new OpenAI();\n const messages: any[] = [{ role: \"system\", content: systemPrompt }, {\n role: \"user\",\n content: typeof userContent === \"string\"\n ? userContent\n : JSON.stringify(userContent),\n }];\n const requestPayload: any = { model, messages, stream };\n if (isJson) requestPayload.response_format = { type: \"json_object\" };\n try {\n const completion = await openai.chat.completions.create(requestPayload);\n if (stream && c) {\n return streamText(c, async (s) => {\n for await (const part of completion) {\n await s.write(part.choices[0]?.delta?.content || \"\");\n }\n });\n }\n if (isJson && c) {\n return c.json(JSON.parse(completion.choices[0].message.content));\n }\n return completion;\n } catch (e) {\n console.error(`Error calling OpenAI: ${e.message}`);\n throw new Error(\"AI service failed.\");\n }\n },\n};\n\nasync function handleExecute(c: Context) {\n try {\n const body: ExecuteBody = await c.req.json();\n const { refined_prompt, user_inputs, company_context, category } = body;\n\n let final_user_prompt = refined_prompt;\n for (const key in user_inputs) {\n const regex = new RegExp(`{{${key}}}`, \"g\");\n final_user_prompt = final_user_prompt.replace(regex, user_inputs[key]);\n }\n\n switch (category) {\n case \"thinking\":\n return services.callOpenAI(\n final_user_prompt,\n \"Execute the task based on the system prompt. Proceed.\",\n { model: config.models.major, stream: true, c },\n );\n\n case \"prompt_building\":\n return services.callOpenAI(\n config.prompts.MASTER_PROMPT_ENGINEER,\n final_user_prompt,\n { model: config.models.major, stream: true, c },\n );\n\n case \"document_creation\":\n const draftCompletion = await services.callOpenAI(\n final_user_prompt,\n \"Execute the task based on the system prompt. Return only the deliverable.\",\n { model: config.models.major },\n );\n const first_draft = draftCompletion.choices[0].message.content;\n\n const criteriaSystemPrompt = config.prompts.SUCCESS_CRITERIA_GENERATOR\n .replace(\"{{task_briefing}}\", final_user_prompt);\n\n const criteriaCompletion = await services.callOpenAI(\n criteriaSystemPrompt,\n \"Generate.\",\n { model: config.models.minor, isJson: true },\n );\n const criteria_json = JSON.parse(\n criteriaCompletion.choices[0].message.content,\n );\n const criteria = criteria_json.criteria;\n\n const evalUserContent = {\n task_briefing: final_user_prompt,\n success_criteria: criteria,\n ai_output: first_draft,\n language: company_context.language,\n };\n\n const evalCompletion = await services.callOpenAI(\n config.prompts.EVALUATOR_AGENT,\n evalUserContent,\n { model: config.models.minor, isJson: true },\n );\n const critique_json = JSON.parse(\n evalCompletion.choices[0].message.content,\n );\n const critique = critique_json.critique;\n\n const refineUserContent = {\n task_briefing: final_user_prompt,\n first_draft: first_draft,\n critique: critique,\n language: company_context.language,\n };\n\n return services.callOpenAI(\n config.prompts.REFINER_AGENT,\n refineUserContent,\n { model: config.models.major, stream: true, c },\n );\n\n default:\n return c.json({ error: \"Invalid task category\" }, 400);\n }\n } catch (e) {\n console.error(`Error in handleExecute: ${e.message}`);\n return c.json({ error: \"Task execution failed.\", details: e.message }, 500);\n }\n}\n\nfunction generateHtml(sourceUrl: string): string {\n const languages = {\n \"English\": \"en-US\",\n \"Spanish\": \"es-ES\",\n \"French\": \"fr-FR\",\n \"German\": \"de-DE\",\n \"Japanese\": \"ja-JP\",\n \"Chinese (Simplified)\": \"zh-CN\",\n };\n const languageOptions = Object.entries(languages).map(([name, code]) =>\n ``\n ).join(\"\");\n const taskCategoriesJson = JSON.stringify(taskCategories);\n return `Taskflow AI | Intelligent Task Automation\n \n \n \n \n \n \n \n \n \n

\n Great job!\n

\n You're making excellent progress.\n

\n \n \n \n

\n \n \n \n \n

\n You're doing amazing! ✨\n

\n Taskflow AI is free and open-source. If you find it valuable,\n consider supporting its development. Your generosity helps\n keep this tool free for everyone!\n

\n ☕ Buy me a coffee

\n This won't show again for 7 days\n

\n \n \n \n

\n \n Welcome to Taskflow AI\n \n

\n Let's get you set up in just a few moments\n

\n \n

\n What makes Taskflow AI special?\n

\n \n AI agents tailored to your industry and needs\n
\n \n Automatic quality refinement for professional\n results\n
\n \n Conversational interface - just answer a few\n questions\n

\n \n

Quick Setup

\n Step 1 of 2\n

\n \n \n

\n Almost there!\n

\n Step 2 of 2\n

\n What best describes your primary use case?\n

\n \n

\n 📣 Marketing & Content\n

\n Create campaigns, copy, and content\n

\n ⚙️ Operations & Strategy\n

\n Plans, processes, and documentation\n

\n 💻 Product & Development\n

\n Technical documentation, specs, and prompts\n

\n ✨ General Business Tasks\n

\n Various professional documents and analyses\n

\n \n

\n \n \n \n \n \n \n \n

\n You're all set! 🎉\n

\n Your workspace is ready. Let's create something amazing\n together!\n

\n \n \n \n \n

\n \n \n \n \n \n \n \n \n TASKFLOW\n \n \n \n \n \n \n \n

\n \n Your Stats\n

\n \n \n 0\n

Tasks

\n \n \n \n 0\n \n

\n Refined\n

\n \n \n \n

\n \n Support Taskflow\n \n

\n \n

\n \n \n \n Workspace Settings\n \n

\n \n

\n \n \n \n Taskflow AI\n \n \n \n

Diff · v88