AI Business Assistant

Thinking

Results

Logs

`;\n}\n\nfunction clientJs(): string {\n return `\nconst msgsDiv = document.getElementById('msgs');\nconst inputArea = document.getElementById('input-area');\nconst loadingInd = document.getElementById('loading');\nconst errDisp = document.getElementById('err-disp');\nconst resDisp = document.getElementById('results-disp');\nconst resCont = document.getElementById('results-cont');\nconst logsCont = document.getElementById('logs-cont');\nlet curSid = null; let curTask = null; let curQ = null; let curSrvState = null;\n\nfunction addMsg(txt, sender) {\n const el = document.createElement('div');\n el.className = 'msg ' + sender + ' msg-appear';\n txt = escHtml(txt).replace(/\\\\[(.*?)\\\\]\\\$(.*?)\\\$/g, '$1');\n el.innerHTML = txt;\n msgsDiv.appendChild(el);\n setTimeout(() => { msgsDiv.scrollTop = msgsDiv.scrollHeight; }, 50);\n}\nfunction addBotMsg(txt) { addMsg(txt, 'bot'); }\nfunction addUserMsg(txt) { addMsg(txt, 'user'); }\nfunction clearInput() { inputArea.innerHTML = ''; }\nfunction enableInput(enabled) { inputArea.querySelectorAll('input, select, textarea, button').forEach(el => el.disabled = !enabled); }\nfunction setLoading(isLoading) { loadingInd.style.display = isLoading ? 'block' : 'none'; }\nfunction showErr(msg) { errDisp.textContent = msg; errDisp.style.display = 'block'; }\nfunction clearErr() { errDisp.textContent = ''; errDisp.style.display = 'none'; }\nfunction escHtml(unsafe) {\n if (unsafe == null) return '';\n let str = unsafe.toString();\n const map = { '&': '&', '<': '<', '>': '>', '\"': '"', \"'\": ''' };\n return str.replace(/[&<>\\\\\"']/g, (m) => map[m]);\n}\n\nfunction showTaskSelect() {\n clearInput();\n if (!curSid || resDisp.style.display === 'none') { addBotMsg(\"Hello! Please select a task:\"); }\n else { addBotMsg(\"What else can I help you with?\"); }\n enableInput(false);\n const sel = document.createElement('select');\n sel.id = 'task-select';\n sel.className = 'fc';\n sel.innerHTML = \\`\n \n \n \n \n \n \n \n \\`;\n const btn = document.createElement('button');\n btn.textContent = 'Start Task';\n btn.onclick = handleTaskSelect;\n inputArea.appendChild(sel);\n inputArea.appendChild(btn);\n enableInput(true);\n sel.focus();\n}\n\nasync function handleTaskSelect() {\n const sel = document.getElementById('task-select');\n const task = sel.value;\n if (!task) { showErr(\"Please select a task.\"); return; }\n clearErr();\n curTask = task;\n addUserMsg(\\`Selected: \\${sel.options[sel.selectedIndex].text}\\`);\n clearInput();\n enableInput(false);\n setLoading(true);\n resDisp.style.display = 'none';\n resDisp.classList.remove('visible');\n logsCont.innerHTML = '';\n resCont.innerHTML = '';\n await sendToServer({ task: curTask, sessionId: curSid, lastAnswer: null, state: curSrvState });\n}\n\nfunction setupInputForQ(q) {\n clearInput();\n curQ = q;\n let inputEl;\n if (q.type === 'select') {\n inputEl = document.createElement('select'); inputEl.id = 'user-input'; inputEl.className = 'fc';\n if (!q.opts || !q.opts.find(opt => opt.val === '')) {\n const defOpt = document.createElement('option'); defOpt.value = ''; defOpt.textContent = '-- Select --'; inputEl.appendChild(defOpt);\n }\n (q.opts || []).forEach(opt => {\n if (opt.val !== '') { const option = document.createElement('option'); option.value = opt.val; option.textContent = opt.lbl; inputEl.appendChild(option); }\n });\n } else if (q.type === 'textarea') {\n inputEl = document.createElement('textarea'); inputEl.id = 'user-input'; inputEl.rows = 3;\n inputEl.placeholder = q.text.includes(\"Optional\") ? \"Your answer (Optional)...\" : \"Your answer...\";\n inputEl.className = 'fc';\n inputEl.addEventListener('keypress', (ev) => { if (ev.key === 'Enter' && !ev.shiftKey) { ev.preventDefault(); handleSend(); } });\n } else {\n inputEl = document.createElement('input'); inputEl.type = 'text'; inputEl.id = 'user-input';\n inputEl.placeholder = q.text.includes(\"Optional\") ? \"Your answer (Optional)...\" : \"Your answer...\";\n inputEl.className = 'fc';\n inputEl.addEventListener('keypress', (ev) => { if (ev.key === 'Enter') { ev.preventDefault(); handleSend(); } });\n }\n const sendBtn = document.createElement('button'); sendBtn.textContent = 'Send'; sendBtn.onclick = handleSend;\n inputArea.appendChild(inputEl);\n inputArea.appendChild(sendBtn);\n enableInput(true);\n inputEl.focus();\n}\n\nasync function handleSend() {\n const inputEl = document.getElementById('user-input');\n if (!inputEl || !curQ) return;\n const ansVal = inputEl.value;\n clearErr();\n let dispVal = ansVal;\n if (curQ.type === 'select') {\n const selOpt = Array.from(inputEl.options).find(opt => opt.value === ansVal);\n dispVal = selOpt ? selOpt.text : (ansVal === \"\" ? \"(Skipped Optional)\" : ansVal);\n } else if (!ansVal.trim() && curQ.text.includes(\"Optional\")) {\n dispVal = \"(Skipped Optional)\";\n } else if (!ansVal.trim()) {\n dispVal = \"(Provided empty input)\";\n }\n if (dispVal !== \"(Provided empty input)\") { addUserMsg(dispVal); }\n else { showErr(\"This field seems required.\"); addUserMsg(dispVal); }\n const lastAns = { field: curQ.field, value: ansVal };\n clearInput();\n enableInput(false);\n setLoading(true);\n await sendToServer({ task: curTask, sessionId: curSid, lastAnswer: lastAns, state: curSrvState });\n}\n\nasync function sendToServer(payload) {\n try {\n console.log(\"Sending to /questionnaire:\", payload);\n const resp = await fetch('/questionnaire', { method: 'POST', headers: { 'Content-Type': 'application/json' }, body: JSON.stringify(payload) });\n setLoading(false); clearErr();\n if (!resp.ok) {\n let errMsg = \\`Server error: \\${resp.status} \\${resp.statusText}\\`;\n try { const errData = await resp.json(); errMsg += \\` - \\${errData.err || errData.message || JSON.stringify(errData)}\\`; } catch(e) {}\n throw new Error(errMsg);\n }\n const data = await resp.json();\n console.log(\"Received from /questionnaire:\", data);\n curSid = data.sid; curSrvState = data.state;\n if (data.err) {\n showErr(data.err);\n if (curQ) { addBotMsg(\\`Error: \\${data.err}. Try again.\\`); setupInputForQ(curQ); }\n else { addBotMsg(\\`Error: \\${data.err}. Select task again.\\`); showTaskSelect(); }\n return;\n }\n if (data.isDone) {\n addBotMsg(\"Got it. Processing...\"); enableInput(false); clearInput(); setLoading(true);\n await execWorkflow(data.data);\n } else if (data.nextQ) {\n addBotMsg(data.nextQ.text); setupInputForQ(data.nextQ);\n } else {\n showErr(\"Unexpected server response.\"); enableInput(false); addBotMsg(\"Server issue. Refresh.\");\n }\n } catch (err) {\n console.error(\"Comm error with /questionnaire:\", err); setLoading(false);\n showErr(\\`Comm error: \\${err.message}. Check connection & refresh.\\`); enableInput(false); addBotMsg(\"Connection problem. Refresh.\");\n }\n}\n\nasync function execWorkflow(finalPayload) {\n enableInput(false);\n try {\n console.log(\"Sending to /execute:\", finalPayload);\n if (!finalPayload || !finalPayload.task || !finalPayload.data) { throw new Error(\"Invalid data for execution.\"); }\n const resp = await fetch('/execute', { method: 'POST', headers: { 'Content-Type': 'application/json' }, body: JSON.stringify(finalPayload) });\n setLoading(false);\n const resData = await resp.json();\n console.log(\"Exec Result:\", resData);\n resDisp.style.display = 'block';\n setTimeout(() => { resDisp.classList.add('visible'); }, 10);\n resCont.innerHTML = ''; logsCont.innerHTML = '';\n if (!resp.ok || resData.err) {\n const errMsg = resData.err || \\`Exec failed: \\${resp.status}\\`;\n resCont.innerHTML = \\`

Error: \\${escHtml(errMsg)}

\\`;\n if (resData.details) { resCont.innerHTML += \\`

Details: \\${escHtml(typeof resData.details === 'string' ? resData.details : JSON.stringify(resData.details))}

\\`; }\n addBotMsg(\\`Error processing: \\${escHtml(errMsg)}\\`);\n } else {\n resCont.innerHTML = formatResults(resData.res, finalPayload.task);\n addBotMsg(\"Request complete! Results below. What else?\");\n }\n logsCont.innerHTML = formatLogs(resData.dbgLog);\n clearInput(); showTaskSelect();\n } catch (err) {\n console.error(\"Comm/Exec error with /execute:\", err); setLoading(false);\n resDisp.style.display = 'block'; setTimeout(() => { resDisp.classList.add('visible'); }, 10);\n resCont.innerHTML = \\`

Critical Exec Error: \\${escHtml(err.message)}

\\`;\n logsCont.innerHTML = formatLogs(null);\n addBotMsg(\"Critical error during processing. Refresh.\"); clearInput(); showTaskSelect();\n }\n}\n\nfunction formatResults(res, task) {\n let html = ''; if (!res) { return '

No specific content generated.

'; }\n switch (task) {\n case \"create_content\":\n html += \\`

Generated Content:

\\${escHtml(res.content || 'N/A')}

\\`;\n if (res.suggestions) html += \\`

Suggestions:

\\${escHtml(res.suggestions)}

\\`;\n break;\n case \"develop_strategy\":\n html += \\`

Marketing Strategy:

\\${escHtml(res.strategy || 'N/A')}

\\`;\n if (res.adIdeas && res.adIdeas.length > 0) html += \\`

Ad Ideas:

\\${escHtml(i)}

\\`;\n if (res.audSugs && res.audSugs.length > 0) html += \\`

Audience Suggestions:

\\${escHtml(s)}

\\`;\n break;\n case \"generate_logo_concept\":\n html += '

Logo Concepts & Descriptions:

';\n if (res.concepts && res.concepts.length > 0) { res.concepts.forEach((c, i) => { html += \\`

Concept \\${i + 1}

\\${escHtml(c)}

\\`; }); }\n else { html += '

No logo concepts generated.

'; }\n if (res.prompts && res.prompts.length > 0) html += \\`

Prompts for Image Models:

```
\\${escHtml(p)}
```

\\`;\n if (res.svg) html += \\`

Basic SVG Attempt:

\\${escHtml(res.svg)}

(Note: SVG quality varies. Save code as '.svg' file.)

\\`;\n html += '

Note: Conceptual starting points.

';\n break;\n case \"build_brand_identity\":\n html += \\`

Brand Voice:

\\${escHtml(res.bVoice || 'N/A')}

\\`;\n if (res.kMsgs && res.kMsgs.length > 0) html += \\`

Key Messages:

\\${escHtml(m)}

\\`;\n html += \\`

Color Suggestions:

\\${escHtml(res.cPalSugs || 'N/A')}

\\`;\n html += \\`

Typography Suggestions:

\\${escHtml(res.typo || 'N/A')}

\\`;\n break;\n case \"run_simulation\":\n html += \\`

Simulation Result:

\\${escHtml(res.simResult || 'N/A')}

\\`;\n if (res.analysis) html += \\`

Analysis:

\\${escHtml(res.analysis)}

\\`;\n break;\n case \"meta_prompt\":\n html += \\`

Generated System Prompt:

\\${escHtml(res.sysPrompt || 'N/A')}

\\`;\n break;\n default: html += \\`

Raw Results:

\\${escHtml(JSON.stringify(res, null, 2))}

\\`;\n } return html;\n}\n\nfunction formatLogs(logs) {\n if (!logs || !Array.isArray(logs) || logs.length === 0) { return '

No logs available.

'; }\n return logs.map(lg => {\n let dtStr = '';\n if (lg.dt != null) {\n try { dtStr = escHtml(JSON.stringify(lg.dt, null, 2)); } catch (e) { dtStr = escHtml(String(lg.dt)); }\n dtStr = \\`

Details:\\\\n\\${dtStr}

\\`;\n }\n const ts = lg.ts ? new Date(lg.ts).toLocaleTimeString() : '??:??:??';\n const lvl = lg.lvl || 'INFO'; const comp = lg.comp || 'System'; const msg = lg.msg || '';\n return \\`

[\\${ts}][\\${lvl}][\\${comp}]\\${escHtml(msg)}\\${dtStr}

\\`;\n }).join('');\n}\nshowTaskSelect(); // Initialize\n`;\n}\n\nclass LogAgent {\n logs: LogEntry[];\n midCtx?: string;\n\n constructor(mid?: string) {\n this.logs = [];\n this.midCtx = mid;\n }\n\n log(lvl: Lvl, comp: string, msg: string, dt?: any, mid?: string, tid?: string): void {\n const effMid = mid || this.midCtx || \"N/A\";\n let safeDt: any;\n if (dt != null) {\n try { safeDt = JSON.parse(JSON.stringify(dt, this.getCircReplacer())); }\n catch (e) { console.error(`[LogAgent] Failed stringify details: ${(e as Error).message}`); safeDt = `[Logging Error: ${(e as Error).message}]`; }\n } else { safeDt = dt; }\n const entry: LogEntry = { ts: new Date().toISOString(), mid: effMid, tid, lvl, comp, msg, dt: safeDt };\n this.logs.push(entry);\n const dtStr = safeDt != null ? `| Details: ${JSON.stringify(safeDt)}` : \"\";\n console.log(`[${entry.ts}] [${entry.mid}] [${entry.tid || \"---\"}] [${entry.lvl}] [${entry.comp}] ${entry.msg} ${dtStr}`);\n }\n\n private getCircReplacer = () => {\n const seen = new WeakSet();\n return (_key: string, value: any) => {\n if (typeof value === \"object\" && value !== null) {\n if (seen.has(value)) { return \"[Circular Reference]\"; }\n seen.add(value);\n }\n return value;\n };\n };\n\n getLogs(): LogEntry[] {\n try { return JSON.parse(JSON.stringify(this.logs)); }\n catch (e) { console.error(\"[LogAgent] Failed deep copy logs:\", e); return [...this.logs]; }\n }\n\n clearLogs(): void { this.logs = []; }\n setMidCtx(mid: string): void { this.midCtx = mid; }\n}\n\nasync function callOpenAI(sysPrompt: string, userPrompt: string, mid: string, tid: string, lg: LogFn): Promise {\n lg(\"DEBUG\", \"callOpenAI\", `Initiating OpenAI call tid=${tid}`, { spLen: sysPrompt.length, upLen: userPrompt.length }, mid, tid);\n try {\n if (!process.env.OPENAI_API_KEY) {\n lg(\"ERROR\", \"callOpenAI\", \"OpenAI API key missing.\", undefined, mid, tid);\n throw new Error(\"OpenAI API key not configured.\");\n }\n const openai = new OpenAI();\n const completion = await openai.chat.completions.create({\n model: \"gpt-4o-mini\",\n messages: [{ role: \"system\", content: sysPrompt }, { role: \"user\", content: userPrompt }],\n temperature: 0.7, max_tokens: 2048,\n });\n const resText = completion.choices[0]?.message?.content;\n const usage = completion.usage;\n if (!resText) {\n lg(\"WARN\", \"callOpenAI\", `OpenAI call returned no text tid=${tid}.`, { usage: usage, finishReason: completion.choices[0]?.finish_reason }, mid, tid);\n return null;\n }\n lg(\"INFO\", \"callOpenAI\", `OpenAI call OK tid=${tid}`, { resLen: resText.length, usage: usage, finishReason: completion.choices[0]?.finish_reason }, mid, tid);\n return resText.trim();\n } catch (err: any) {\n const errDtls = { msg: err.message, name: err.name, status: err.status, type: err.type, code: err.code };\n lg(\"ERROR\", \"callOpenAI\", `OpenAI API call failed tid=${tid}`, { error: errDtls }, mid, tid);\n throw new Error(`OpenAI API failed: ${err.message}` + (err.code ? ` (Code: ${err.code})` : ''));\n }\n}\n\nclass QAgent {\n lg: LogFn;\n private baseReq = [\"bType\", \"sDesc\", \"tAud\"];\n private baseOpt = [\"bTone\", \"goals\"];\n private taskSpec: Record = {\n \"create_content\": { req: [\"cType\", \"plat\"], opt: [\"specInstr\"] },\n \"develop_strategy\": { req: [\"plats\"], opt: [\"budgetLvl\", \"duration\"] },\n \"generate_logo_concept\": { req: [\"bName\"], opt: [\"stylePrefs\", \"colorPrefs\"] },\n \"build_brand_identity\": { req: [\"bName\"], opt: [\"coreVals\"] },\n \"run_simulation\": { req: [\"simType\", \"specParams\"], opt: [] },\n \"meta_prompt\": { req: [\"pInput\"], opt: [] },\n };\n\n constructor(lg: LogFn) { this.lg = lg; }\n\n private getReqFlds(task: UserIn[\"task\"]): string[] {\n if (task === \"meta_prompt\") { return this.taskSpec.meta_prompt.req; }\n const req = [...this.baseReq, ...(this.taskSpec[task]?.req || [])];\n const opt = [...this.baseOpt, ...(this.taskSpec[task]?.opt || [])];\n return [...new Set([...req, ...opt])];\n }\n\n private isOptional(field: string, task: UserIn[\"task\"]): boolean {\n if (task === \"meta_prompt\") return false;\n const baseOpt = this.baseOpt.includes(field);\n const taskOpt = (this.taskSpec[task]?.opt || []).includes(field);\n return baseOpt || taskOpt;\n }\n\n private selectNextField(fields: string[], answers: Record, task: UserIn[\"task\"]): string | null {\n for (const fld of fields) {\n const isMissing = !(fld in answers);\n const isEmptyReq = (fld in answers) && (answers[fld] == null || answers[fld] === '') && !this.isOptional(fld, task);\n if (isMissing || isEmptyReq) { return fld; }\n }\n return null;\n }\n\n private formatQuest(field: string, curState: QState): Quest {\n const qid = `${field}-${Date.now()}`;\n let text = `What is your ${field.replace(/([A-Z])/g, ' $1').toLowerCase()}?`;\n let type: Quest[\"type\"] = \"text\";\n let opts: Quest[\"opts\"] | undefined = undefined;\n const bName = curState.answers.bName || \"your business\";\n const isOpt = this.isOptional(field, curState.task);\n const optSuffix = isOpt ? \" (Optional)\" : \"\";\n\n switch (field) {\n case \"bType\": text = `Type of business for ${bName}? (e.g., SaaS, Bakery)`; break;\n case \"sDesc\": text = `Describe products/services for ${bName}.`; type = \"textarea\"; break;\n case \"tAud\": text = `Ideal customer for ${bName}? (Specific, e.g., Tech startups)`; type = \"textarea\"; break;\n case \"bTone\": text = `Desired tone for ${bName}?${optSuffix}`; type = \"select\"; opts = [{ lbl: `-- Tone${optSuffix} --`, val: \"\" }, { lbl: \"Professional\", val: \"professional\" }, { lbl: \"Friendly\", val: \"friendly\" }, { lbl: \"Casual\", val: \"casual\" }, { lbl: \"Humorous\", val: \"humorous\" }, { lbl: \"Informative\", val: \"informative\" }]; break;\n case \"goals\": text = `Main goals for this task?${optSuffix}`; type = \"textarea\"; break;\n case \"bName\": text = `Exact business name?`; break;\n case \"cType\": text = \"Content type needed?\"; type = \"select\"; opts = [{ lbl: \"-- Type --\", val: \"\" }, { lbl: \"Promo Post\", val: \"promo_post\" }, { lbl: \"Blog Ideas\", val: \"blog_idea\" }, { lbl: \"Blog Draft\", val: \"blog_draft\" }, { lbl: \"Ad Copy\", val: \"ad_copy\" }]; break;\n case \"plat\": text = \"Platform for content?\"; type = \"select\"; opts = [{ lbl: \"-- Platform --\", val: \"\" }, { lbl: \"Generic\", val: \"generic\" }, { lbl: \"Facebook\", val: \"facebook\" }, { lbl: \"Instagram\", val: \"instagram\" }, { lbl: \"LinkedIn\", val: \"linkedin\" }, { lbl: \"Blog\", val: \"blog\" }]; break;\n case \"specInstr\": text = `Specific instructions/keywords?${optSuffix}`; type = \"textarea\"; break;\n case \"plats\": text = \"Strategy platforms? (e.g., LinkedIn, SEO - comma separated)\"; break;\n case \"budgetLvl\": text = `Approx budget level?${optSuffix}`; type = \"select\"; opts = [{ lbl: `-- Budget${optSuffix} --`, val: \"\" }, { lbl: \"Low (Organic)\", val: \"low\" }, { lbl: \"Medium (Mod Ads)\", val: \"medium\" }, { lbl: \"High (Sig Ads)\", val: \"high\" }]; break;\n case \"duration\": text = `Strategy duration?${optSuffix}`; break;\n case \"stylePrefs\": text = `Logo style preferences?${optSuffix}`; break;\n case \"colorPrefs\": text = `Logo color preferences?${optSuffix}`; break;\n case \"coreVals\": text = `Core values/mission of ${bName}?${optSuffix}`; type = \"textarea\"; break;\n case \"simType\": text = \"Simulation type?\"; type = \"select\"; opts = [{ lbl: \"-- Sim --\", val: \"\" }, { lbl: \"Ad Response Est.\", val: \"ad_response_estimate\" }, { lbl: \"Customer Interaction\", val: \"customer_interaction\" }, { lbl: \"Other\", val: \"other\" }]; break;\n case \"specParams\": text = `Sim details/params? (Ad: audience, platform, offer. Cust: scenario, issue. Other: describe.)`; type = \"textarea\"; break;\n case \"pInput\": text = \"AI task desc or prompt to refine?\"; type = \"textarea\"; break;\n default: text = `Please provide '${field}'.${optSuffix}`; if (field.toLowerCase().includes(\"desc\") || field.toLowerCase().includes(\"details\")) { type = \"textarea\"; } break;\n }\n if (isOpt && !text.includes(\"(Optional)\")) { text += \" (Optional)\"; }\n return { id: qid, field, text, type, opts };\n }\n\n async nextStep(curSid: string | null, task: UserIn[\"task\"], lastAns: { field: string; value: any } | null, curSt?: QState | null): Promise {\n let state: QState;\n const sid = curSid || `sess-${Date.now()}-${Math.random().toString(36).substring(2, 8)}`;\n\n if (!curSt || !curSid || curSt.task !== task) {\n this.lg(\"INFO\", \"QAgent\", `Init/Reset state for task: ${task}`, { oldSid: curSid, oldTask: curSt?.task }, sid);\n const reqFlds = this.getReqFlds(task);\n let initAns: Partial> = {};\n if (curSt && curSt.sid === sid && curSt.task !== task) {\n this.lg(\"INFO\", \"QAgent\", `Task changed ${curSt.task} -> ${task}. Retaining common.`, undefined, sid);\n const common = [\"bType\", \"sDesc\", \"tAud\", \"bTone\", \"goals\", \"bName\"];\n common.forEach(f => { if (f in curSt.answers && curSt.answers[f] != null && curSt.answers[f] !== '') { initAns[f] = curSt.answers[f]; } });\n this.lg(\"DEBUG\", \"QAgent\", `Retained answers:`, initAns, sid);\n }\n state = { sid: sid, task: task, reqFlds: reqFlds, answers: initAns, qHist: [], isDone: false };\n if (lastAns?.field && lastAns.value !== undefined) {\n this.lg(\"DEBUG\", \"QAgent\", `Applying init answer post-reset: Field '${lastAns.field}'`, { val: lastAns.value }, sid);\n const val = typeof lastAns.value === 'string' ? lastAns.value.trim() : lastAns.value;\n if (val !== '' || this.isOptional(lastAns.field, state.task)) { state.answers[lastAns.field] = val; }\n }\n } else {\n state = { ...curSt, answers: { ...curSt.answers }, qHist: [...curSt.qHist] };\n if (lastAns?.field && lastAns.value !== undefined) {\n this.lg(\"DEBUG\", \"QAgent\", `Processing answer for field '${lastAns.field}'`, { val: lastAns.value }, state.sid);\n const val = typeof lastAns.value === 'string' ? lastAns.value.trim() : lastAns.value;\n if (val !== '' || this.isOptional(lastAns.field, state.task)) { state.answers[lastAns.field] = val; }\n else { this.lg(\"WARN\", \"QAgent\", `Received empty value for required field '${lastAns.field}'. Storing anyway.`, undefined, state.sid); state.answers[lastAns.field] = val; }\n }\n }\n\n const nextFld = this.selectNextField(state.reqFlds, state.answers, state.task);\n\n if (!nextFld) {\n this.lg(\"INFO\", \"QAgent\", `All fields gathered for task '${state.task}'. Validating.`, { gathered: state.answers }, state.sid);\n const finalData = this.finalizeData(state.answers, state.task, state.sid);\n try {\n UserInSchema.parse(finalData);\n this.lg(\"INFO\", \"QAgent\", \"Final data validation OK.\", { finalData }, state.sid);\n state.isDone = true;\n return { sid: state.sid, isDone: true, data: finalData, state, nextQ: undefined, err: undefined };\n } catch (validErr) {\n if (validErr instanceof z.ZodError) {\n const errMsgs = validErr.issues.map(i => `Field '${i.path.join('.')}': ${i.message}`).join('; ');\n this.lg(\"ERROR\", \"QAgent\", \"Final data validation FAILED.\", { errs: validErr.issues, dataAttempted: finalData }, state.sid);\n state.isDone = false;\n const firstFailField = validErr.issues[0]?.path[1]?.toString();\n if (firstFailField && state.reqFlds.includes(firstFailField)) {\n this.lg(\"WARN\", \"QAgent\", `Validation failed on field '${firstFailField}'. Re-asking question.`, { error: errMsgs }, state.sid);\n const nextQ = this.formatQuest(firstFailField, state);\n if (!state.qHist.find(q => q.qid === nextQ.id)) { state.qHist.push({ qid: nextQ.id, field: nextQ.field, text: nextQ.text }); }\n return { sid: state.sid, isDone: false, nextQ, state, data: undefined, err: `Validation Error: ${errMsgs}` };\n } else {\n this.lg(\"ERROR\", \"QAgent\", \"Validation failed, couldn't identify field to re-ask.\", { error: errMsgs, issues: validErr.issues }, state.sid);\n return { sid: state.sid, isDone: false, err: `Data validation failed: ${errMsgs}. Cannot proceed.`, state, nextQ: undefined, data: undefined };\n }\n } else {\n this.lg(\"ERROR\", \"QAgent\", \"Unexpected error during final data validation.\", { error: (validErr as Error).message, dataAttempted: finalData }, state.sid);\n state.isDone = false;\n return { sid: state.sid, isDone: false, err: `Unexpected validation error: ${(validErr as Error).message}`, state, nextQ: undefined, data: undefined };\n }\n }\n } else {\n const nextQ = this.formatQuest(nextFld, state);\n if (!state.qHist.find(q => q.qid === nextQ.id)) { state.qHist.push({ qid: nextQ.id, field: nextQ.field, text: nextQ.text }); }\n this.lg(\"INFO\", \"QAgent\", `Asking next question for field: '${nextFld}'`, { qText: nextQ.text }, state.sid);\n return { sid: state.sid, isDone: false, nextQ, state, data: undefined, err: undefined };\n }\n }\n\n private finalizeData(gathered: Record, task: UserIn[\"task\"], sid?: string): UserIn {\n let finalPayloadData: Record = { ...gathered };\n const allFields = this.getReqFlds(task);\n allFields.forEach(field => {\n if (this.isOptional(field, task) && (finalPayloadData[field] == null || finalPayloadData[field] === \"\")) {\n delete finalPayloadData[field];\n this.lg(\"DEBUG\", \"finalizeData\", `Removed empty optional field: ${field}`, undefined, sid);\n }\n });\n\n if (task === 'develop_strategy' && finalPayloadData.plats !== undefined) {\n if (typeof finalPayloadData.plats === 'string') {\n const raw = finalPayloadData.plats;\n finalPayloadData.plats = raw.split(',').map((p: string) => p.trim().toLowerCase()).filter((p: string) => p.length > 0);\n this.lg(\"DEBUG\", \"finalizeData\", `Transformed plats string to array`, { orig: raw, res: finalPayloadData.plats }, sid);\n } else if (!Array.isArray(finalPayloadData.plats)) {\n this.lg(\"WARN\", \"finalizeData\", `Plats field not string/array. Setting empty.`, { val: finalPayloadData.plats }, sid);\n finalPayloadData.plats = [];\n }\n finalPayloadData.plats = (finalPayloadData.plats as any[]).filter((p: any) => typeof p === 'string');\n }\n\n if (task === 'run_simulation' && finalPayloadData.specParams !== undefined) {\n if (typeof finalPayloadData.specParams === 'string') {\n const paramStr = finalPayloadData.specParams.trim();\n if (paramStr.startsWith('{') && paramStr.endsWith('}')) {\n try {\n const parsed = JSON.parse(paramStr);\n if (typeof parsed === 'object' && parsed !== null && !Array.isArray(parsed)) { finalPayloadData.specParams = parsed; this.lg(\"DEBUG\", \"finalizeData\", `Parsed specParams string as JSON object.`, { parsed }, sid); }\n else { finalPayloadData.specParams = { description: paramStr }; this.lg(\"WARN\", \"finalizeData\", `Sim specParams non-object JSON, wrapping.`, { raw: gathered.specParams }, sid); }\n } catch (e) { finalPayloadData.specParams = { description: paramStr }; this.lg(\"WARN\", \"finalizeData\", `Sim specParams invalid JSON, wrapping.`, { raw: gathered.specParams, err: (e as Error).message }, sid); }\n } else { finalPayloadData.specParams = { description: paramStr }; this.lg(\"DEBUG\", \"finalizeData\", `Sim specParams string treated as description.`, { desc: paramStr }, sid); }\n } else if (typeof finalPayloadData.specParams !== 'object' || finalPayloadData.specParams === null || Array.isArray(finalPayloadData.specParams)) {\n this.lg(\"WARN\", \"finalizeData\", `Sim specParams not string/object. Wrapping.`, { val: finalPayloadData.specParams }, sid);\n finalPayloadData.specParams = { description: JSON.stringify(finalPayloadData.specParams) };\n }\n }\n // Rename fields before returning to match schema expectations\n const finalSchemaData: Record = {};\n const keyMap: Record = {\n bType: 'businessType', sDesc: 'serviceDescription', tAud: 'targetAudience', bTone: 'brandTone', goals: 'goals',\n cType: 'contentType', plat: 'platform', specInstr: 'specificInstructions', plats: 'platforms', budgetLvl: 'budgetLevel',\n duration: 'duration', bName: 'businessName', stylePrefs: 'stylePreferences', colorPrefs: 'colorPreferences',\n coreVals: 'coreValues', simType: 'simulationType', specParams: 'specificParams', pInput: 'prompt_input'\n };\n for (const key in finalPayloadData) {\n finalSchemaData[keyMap[key] || key] = finalPayloadData[key];\n }\n\n\n return { task: task, data: finalSchemaData };\n }\n}\n\n\nasync function runCntAgent(task: TskIn, lg: LogFn): Promise> {\n const { mid, tid, p } = task;\n lg(\"INFO\", \"CntAgent\", `Starting content gen tid=${tid}`, { cType: p.cType, plat: p.plat, goals: p.goals }, mid, tid);\n const sysP = `Expert copywriter for SMBs. Goal: generate content. CONTEXT: BT: ${p.bType || 'N/A'}, SVC: ${p.sDesc || 'N/A'}, TA: ${p.tAud || 'General'}, Tone: ${p.bTone || 'Pro/Approach'}, Goal: ${p.goals || 'Engage/Promo'}. TASK: Type: ${p.cType}, Platform: ${p.plat}, Instr: ${p.specInstr || 'None'}. INSTR: 1. Gen '${p.cType}' for '${p.plat}'. 2. Use Tone ('${p.bTone || 'Pro/Approach'}') for TA ('${p.tAud || 'General'}'). 3. Integrate Instr: \"${p.specInstr || 'None'}\". 4. Align w/ Goal ('${p.goals || 'Engage/Promo'}'), use CTAs. 5. Platform: Socials(hashtags,length), Ads(headlines,desc,CTA), BlogIdea(3-5 titles), BlogDraft(300-500w section,markdown), Generic(adaptable). 6. Output: ONLY content. Blog Ideas: bullet points.`;\n const userP = `Generate the ${p.cType} for the specified platform based on the context provided in the system prompt.`;\n try {\n const genContent = await callOpenAI(sysP, userP, mid, tid, lg);\n if (!genContent) { lg(\"WARN\", \"CntAgent\", `LLM returned no content tid=${tid}.`, undefined, mid, tid); return { mid, cid: tid, p: { content: \"\" }, e: \"AI failed to generate content.\" }; }\n const output: CntOut = { content: genContent };\n if (p.cType === 'promo_post' || p.cType === 'ad_copy') { output.suggestions = \"Pair with image/video. A/B test CTAs.\"; }\n else if (p.cType === 'blog_idea') { output.suggestions = \"Research keywords. Outline chosen idea.\"; }\n else if (p.cType === 'blog_draft') { output.suggestions = \"Expand draft. Check keywords, add links.\"; }\n lg(\"INFO\", \"CntAgent\", `Content gen task OK tid=${tid}.`, { outLen: output.content.length, sug: !!output.suggestions }, mid, tid);\n return { mid, cid: tid, p: output };\n } catch (err: any) { lg(\"ERROR\", \"CntAgent\", `Content gen task fail tid=${tid}.`, { err: err.message }, mid, tid); return { mid, cid: tid, p: { content: \"\" }, e: `Content generation failed: ${err.message}` }; }\n}\n\nasync function runStratAgent(task: TskIn, lg: LogFn): Promise> {\n const { mid, tid, p } = task;\n const platsStr = Array.isArray(p.plats) ? p.plats.join(\", \") : \"N/A\";\n lg(\"INFO\", \"StratAgent\", `Starting strategy dev tid=${tid}`, { plats: platsStr, budget: p.budgetLvl, dur: p.duration }, mid, tid);\n const sysP = `Pragmatic Digital Mktg Strategist for SMBs. Task: outline concise strategy. CONTEXT: BT: ${p.bType || 'N/A'}, SVC: ${p.sDesc || 'N/A'}, TA: ${p.tAud || 'General'}, Goals: ${p.goals || 'Incr presence/leads'}, Platforms: ${platsStr}, Budget: ${p.budgetLvl || 'Low'}, Duration: ${p.duration || 'Ongoing'}, Tone: ${p.bTone || 'Pro/Trust'}. FRAMEWORK (Use MD headings): ## Executive Summary - Core strategy/outcomes. ## Target Audience Insights - Online behavior, needs, reach. ## Recommended Platforms & Rationale - Eval requested ('${platsStr}'). Recommend best based on audience/goals/budget. Justify. Suggest alt if needed. ## Core Content Themes & Messaging - 2-4 themes for audience/value. Key messages. ## Platform-Specific Tactics (recommended) - 2-3 tactics per platform (e.g., LinkedIn: insights, groups; SEO: long-tail kw, citations). Prioritize by Budget (Low: organic; Med: ads; High: campaigns). ## Sample Ad Copy / Post Ideas - 2-3 examples. ## Budget Considerations (If applicable) - Med/High: spend priority. Low: acknowledge limits. ## Measurement & KPIs - 3-5 KPIs (e.g., Traffic, CVR, Engagement, CPL). ## Next Steps - Immediate actions (e.g., Profiles, KW research, Calendar). INSTR: 1. Practical, specific. 2. Prioritize by Budget. 3. Cohesive. 4. Output ONLY strategy doc (MD). No intro/outro text.`;\n const userP = `Develop the marketing strategy based on the system prompt context & framework.`;\n try {\n const stratContent = await callOpenAI(sysP, userP, mid, tid, lg);\n if (!stratContent) { lg(\"WARN\", \"StratAgent\", `LLM no content for strat tid=${tid}.`, undefined, mid, tid); return { mid, cid: tid, p: { strategy: \"\" }, e: \"AI failed strategy.\" }; }\n let adIdeas: string[] = [];\n const adMatch = stratContent.match(/## Sample Ad Copy \\/ Post Ideas\\n+([\\s\\S]*?)(?=\\n##|$)/i);\n if (adMatch?.[1]) { adIdeas = adMatch[1].split(/[\\n\\r]+-|\\*|\\d+\\.\\s*/).map(s => s.trim()).filter(s => s && s.length > 10); }\n let audSugs: string[] = [];\n const audMatch = stratContent.match(/## Target Audience Insights\\n+([\\s\\S]*?)(?=\\n##|$)/i);\n if (audMatch?.[1]) { audSugs = audMatch[1].split(/[\\n\\r]+-|\\*|\\d+\\.\\s*/).map(s => s.trim()).filter(s => s && s.length > 10); }\n const output: StratOut = { strategy: stratContent, adIdeas, audSugs };\n lg(\"INFO\", \"StratAgent\", `Strategy dev OK tid=${tid}.`, { stratLen: output.strategy.length, adIdeasCnt: adIdeas.length, audSugsCnt: audSugs.length }, mid, tid);\n return { mid, cid: tid, p: output };\n } catch (err: any) { lg(\"ERROR\", \"StratAgent\", `Strategy dev fail tid=${tid}.`, { err: err.message }, mid, tid); return { mid, cid: tid, p: { strategy: \"\" }, e: `Strategy generation failed: ${err.message}` }; }\n}\n\nasync function runLogoAgent(task: TskIn, lg: LogFn): Promise> {\n const { mid, tid, p } = task;\n lg(\"INFO\", \"LogoAgent\", `Starting logo concept gen tid=${tid}`, { bName: p.bName, style: p.stylePrefs, color: p.colorPrefs }, mid, tid);\n const sysP = `Creative Asst gen 2-3 logo concepts. Each: desc + AI img prompt. CONTEXT: BN: ${p.bName}, BT: ${p.bType || 'N/A'}, Svc: ${p.sDesc || 'N/A'}, TA: ${p.tAud || 'General'}, Styles: ${p.stylePrefs || 'Versatile'}, Colors: ${p.colorPrefs || 'Suggest'}, Tone: ${p.bTone || 'Professional'}. INSTR: 1. Gen 2-3 diff concepts (icon, wordmark, abstract/combo). 2. Format each (MD): ### Concept Title \\\\n#### Description: \\\\n- Visuals, symbolism, fonts (sans/serif), colors. \\\\n#### AI Image Generation Prompt: \\\\n- Detailed prompt for AI (DALL-E, etc). Incl: type(vector logo), subj(BN/initials), style(minimalist, flat), elements, colors, bg(white). Ex: \"/imagine prompt: vector logo for '${p.bName}', minimalist flat design [Symbol], clean sans-serif, blue (#005C99), gray (#CCCCCC), isolated white bg --style raw --ar 1:1\" 3. Separate concepts w/ '---'. 4. (Opt) If simple, attempt basic SVG: \\`\\`\\`svg ... \\`\\`\\`. 5. Output ONLY concepts. No intro/outro.`;\n const userP = `Generate 2-3 distinct logo concepts for \"${p.bName}\" based on the system prompt. Provide descriptions and AI prompts.`;\n try {\n const conceptContent = await callOpenAI(sysP, userP, mid, tid, lg);\n if (!conceptContent) { lg(\"WARN\", \"LogoAgent\", `LLM no content for logo tid=${tid}.`, undefined, mid, tid); return { mid, cid: tid, p: { concepts: [], prompts: [] }, e: \"AI failed logo concepts.\" }; }\n const concepts: string[] = []; const prompts: string[] = [];\n const sections = conceptContent.split(/\\n---\\n/);\n sections.forEach(sec => {\n const titleM = sec.match(/### (.*?)\\n/);\n const descM = sec.match(/#### Description:\\n([\\s\\S]*?)(?=\\n#### AI Image Generation Prompt:|$)/i);\n const promptM = sec.match(/#### AI Image Generation Prompt:\\n([\\s\\S]*?)(?=\\n\\`\\`\\`svg|$)/i);\n if (titleM && descM && promptM) {\n concepts.push(`Title: ${titleM[1].trim()}\\n\\nDescription:\\n${descM[1].trim()}`); prompts.push(promptM[1].trim());\n } else { lg(\"WARN\", \"LogoAgent\", `Could not parse logo concept section.`, { sectionStart: sec.substring(0, 100) }, mid, tid); if (sec.trim().length > 0) concepts.push(`Raw Section:\\n${sec.trim()}`); }\n });\n const svgM = conceptContent.match(/```svg\\n([\\s\\S]*?)\\n```/); const svgAttempt = svgM ? svgM[1].trim() : undefined;\n if (concepts.length === 0 && conceptContent.length > 0) { lg(\"WARN\", \"LogoAgent\", `No structured concepts parsed, return raw tid=${tid}`, { rawLen: conceptContent.length }, mid, tid); concepts.push(`Raw Output (Parse Fail):\\n${conceptContent}`); }\n const output: LogoOut = { concepts, prompts, svg: svgAttempt };\n lg(\"INFO\", \"LogoAgent\", `Logo concept task OK tid=${tid}.`, { concepts: output.concepts.length, prompts: output.prompts?.length, svg: !!output.svg }, mid, tid);\n return { mid, cid: tid, p: output };\n } catch (err: any) { lg(\"ERROR\", \"LogoAgent\", `Logo concept task fail tid=${tid}.`, { err: err.message }, mid, tid); return { mid, cid: tid, p: { concepts: [], prompts: [] }, e: `Logo concept generation failed: ${err.message}` }; }\n}\n\nasync function runBrandAgent(task: TskIn, lg: LogFn): Promise> {\n const { mid, tid, p } = task;\n lg(\"INFO\", \"BrandAgent\", `Starting brand ID def tid=${tid}`, { bName: p.bName, vals: p.coreVals, tone: p.bTone }, mid, tid);\n const sysP = `Brand Strategist Asst defining foundational brand ID. Create concise guide. CONTEXT: BN: ${p.bName}, BT: ${p.bType || 'N/A'}, Svc: ${p.sDesc || 'N/A'}, TA: ${p.tAud || 'General'}, Values: ${p.coreVals || 'N/A'}, Tone: ${p.bTone || 'Pro/Trust'}. FRAMEWORK (MD headings): ## Brand Voice & Personality - Describe voice detail. Elaborate tone ('${p.bTone || 'Pro/Trust'}'). Define personality (Expert, Friendly, Reliable). Lang style (Formal/Casual, Jargon/Plain). ## Key Messages - List 3-5 core msgs/taglines (* bullets). Communicate value, connect service->audience, reflect values. Concise, memorable. ## Color Palette Suggestions - Recommend primary, 1-2 secondary, 1 accent color. Hex codes (#RRGGBB). Justify (personality, industry, audience). Ex: \"Primary: Blue (#0A2F5B) trust. Secondary: Gray (#E1E1E1) contrast. Accent: Teal (#1ABC9C) energy.\" ## Typography Suggestions - Recommend 1 heading, 1 body font (Google Fonts). Justify (readability, personality, contrast). Ex: \"Head: Montserrat Bold (Modern). Body: Lato Regular (Clean). Prof feel.\" INSTR: 1. Cohesive suggestions (Voice, Msgs, Colors, Fonts). 2. Practical for SMB. 3. Align w/ Context (BT, TA, Tone). 4. Output ONLY guide (MD). No intro/outro.`;\n const userP = `Develop foundational brand ID guide for \"${p.bName}\" based on system prompt context/instructions. Ensure cohesive & practical.`;\n try {\n const idContent = await callOpenAI(sysP, userP, mid, tid, lg);\n if (!idContent) { lg(\"WARN\", \"BrandAgent\", `LLM no content for brand ID tid=${tid}.`, undefined, mid, tid); return { mid, cid: tid, p: { bVoice: \"\", kMsgs: [], cPalSugs: \"\", typo: \"\" }, e: \"AI failed brand ID guide.\" }; }\n const voiceM = idContent.match(/## Brand Voice & Personality\\n([\\s\\S]*?)(?=\\n##|$)/i);\n const msgsM = idContent.match(/## Key Messages\\n([\\s\\S]*?)(?=\\n##|$)/i);\n const colorM = idContent.match(/## Color Palette Suggestions\\n([\\s\\S]*?)(?=\\n##|$)/i);\n const typoM = idContent.match(/## Typography Suggestions\\n([\\s\\S]*?)(?=\\n##|$)/i);\n const bVoice = voiceM ? voiceM[1].trim() : \"Parse fail Voice.\";\n const kMsgsRaw = msgsM ? msgsM[1].trim() : \"\";\n let kMsgs = kMsgsRaw.split(/[\\n\\r]+\\s*[\\*\\-]\\s*/).map(s => s.trim()).filter(s => s.length > 5);\n if (kMsgs.length === 0 && kMsgsRaw.length > 5) { kMsgs.push(`Raw Msgs: ${kMsgsRaw}`); }\n const cPalSugs = colorM ? colorM[1].trim() : \"Parse fail Colors.\";\n const typo = typoM ? typoM[1].trim() : \"Parse fail Typography.\";\n if (bVoice.startsWith(\"Parse fail\") && kMsgs.length === 0) { lg(\"WARN\", \"BrandAgent\", `Sig parse fail brand ID tid=${tid}`, { rawP: idContent.substring(0, 200) }, mid, tid); }\n const output: BrandOut = { bVoice, kMsgs, cPalSugs, typo };\n lg(\"INFO\", \"BrandAgent\", `Brand ID task OK tid=${tid}.`, { voiceLen: output.bVoice.length, msgsCnt: output.kMsgs.length }, mid, tid);\n return { mid, cid: tid, p: output };\n } catch (err: any) { lg(\"ERROR\", \"BrandAgent\", `Brand ID task fail tid=${tid}.`, { err: err.message }, mid, tid); return { mid, cid: tid, p: { bVoice: \"\", kMsgs: [], cPalSugs: \"\", typo: \"\" }, e: `Brand ID generation failed: ${err.message}` }; }\n}\n\nasync function runSimAgent(task: TskIn, lg: LogFn): Promise> {\n const { mid, tid, p } = task;\n let paramDesc = \"N/A\";\n try { paramDesc = typeof p.specParams === 'string' ? p.specParams : JSON.stringify(p.specParams); }\n catch (e) { paramDesc = \"[Serialization fail]\"; lg(\"WARN\", \"SimAgent\", `Failed serialize specParams tid=${tid}`, { err: (e as Error).message }, mid, tid); }\n lg(\"INFO\", \"SimAgent\", `Starting sim task tid=${tid}`, { simType: p.simType, paramsP: paramDesc.substring(0, 100) }, mid, tid);\n const sysP = `Adv Sim Model predicts outcomes/role-plays. CONTEXT: BT: ${p.bType || 'N/A'}, Svc: ${p.sDesc || 'N/A'}, TA: ${p.tAud || 'General'}, Tone: ${p.bTone || 'Pro/Helpful'}. SIM TASK: Type: ${p.simType}, Params/Scenario: ${paramDesc}. INSTR: 1. Analyze Type & Params. 2. Execute: 'ad_response_estimate': Qual assess ad effectiveness (platform, audience, copy). Estimate metric ranges (CTR, CVR) if feasible. State assumptions (avg market). Avoid guarantees. Focus likelihoods. 'customer_interaction': Role-play realistic interaction (Customer:, Agent:). Reflect context, tone, issue. Concise (3-5 exchanges). 'other': Execute per desc. Interpret reasonably. State assumptions. 3. Output: Core sim result first. THEN optional \\`## Analysis\\` (MD heading). Brief takeaways/insights from sim only (2-4 sentences). 4. Output ONLY result & opt analysis. No intro/outro.`;\n const userP = `Run simulation type '${p.simType}' based on system prompt context/params. Follow format.`;\n try {\n const simContent = await callOpenAI(sysP, userP, mid, tid, lg);\n if (!simContent) { lg(\"WARN\", \"SimAgent\", `LLM no content for sim tid=${tid}.`, undefined, mid, tid); return { mid, cid: tid, p: { simResult: \"\" }, e: \"AI failed sim.\" }; }\n let simResult = simContent; let analysis: string | undefined = undefined;\n const analysisM = simContent.match(/^## Analysis\\s*$/mi);\n if (analysisM?.index !== undefined) {\n simResult = simContent.substring(0, analysisM.index).trim();\n analysis = simContent.substring(analysisM.index).replace(/^## Analysis\\s*[\\r\\n]+/mi, '').trim();\n lg(\"DEBUG\", \"SimAgent\", `Analysis section found tid=${tid}`, { analysisLen: analysis.length }, mid, tid);\n } else { simResult = simContent.trim(); lg(\"DEBUG\", \"SimAgent\", `No analysis section found tid=${tid}`, undefined, mid, tid); }\n if (!simResult && analysis) { lg(\"WARN\", \"SimAgent\", `Sim result empty, analysis found tid=${tid}.`, undefined, mid, tid); simResult = \"(Simulation produced analysis but no primary result)\"; }\n else if (!simResult && !analysis) { lg(\"WARN\", \"SimAgent\", `Sim result & analysis empty tid=${tid}.`, undefined, mid, tid); simResult = \"(Simulation produced no output)\"; }\n const output: SimOut = { simResult, analysis };\n lg(\"INFO\", \"SimAgent\", `Sim task OK tid=${tid}.`, { resLen: output.simResult.length, hasAnalysis: !!output.analysis }, mid, tid);\n return { mid, cid: tid, p: output };\n } catch (err: any) { lg(\"ERROR\", \"SimAgent\", `Sim task fail tid=${tid}.`, { err: err.message }, mid, tid); return { mid, cid: tid, p: { simResult: \"\" }, e: `Simulation run failed: ${err.message}` }; }\n}\n\nasync function runMetaAgent(task: TskIn, lg: LogFn): Promise> {\n const { mid, tid, p } = task;\n lg(\"INFO\", \"MetaAgent\", `Starting meta-prompt gen/refine tid=${tid}`, { inputLen: p.pInput.length }, mid, tid);\n const sysP = `Expert AI Prompt Engineer. Mission: create/refine high-quality SYSTEM prompt for another AI (GPT-4) based on user's request. Prompt should be clear, comprehensive, unambiguous, elicit best response. PRINCIPLES: 1. Role & Expertise: Define AI persona (e.g., \"Expert financial analyst...\"). 2. Task/Goal: State main objective (e.g., \"...analyze data, gen memo.\"). 3. Context/Input Spec: List needed inputs ([Business Plan], [User Question]). 4. Process/Steps (if complex): Outline steps (1. Analyze. 2. Risks. 3. Recs. 4. Output.). 5. Output Format/Constraints: Specify format(MD, JSON), len, tone, style, language, avoid (\"JSON format.\", \"No filler.\", \"< 500 words.\"). Output structure examples if helpful. 6. Ambiguity/Missing Info: Instruct handling (\"Ask clarifying Qs.\", \"State missing data.\"). 7. Refinement (If Applicable): Analyze existing prompt strengths/weaknesses. Address weaknesses (clarity, format, role). USER'S REQUEST (for sys prompt gen/refine): \\`\\`\\`\\n${p.pInput}\\n\\`\\`\\` YOUR TASK: Generate complete system prompt text based ONLY on user req & principles. Output ONLY system prompt itself. No code blocks unless user requested. No explanations/intros (\"Here is prompt:\").`;\n const userP = `Based on my request in system context, generate/refine the system prompt per guidelines. Output only resulting prompt text.`;\n try {\n const genSysP = await callOpenAI(sysP, userP, mid, tid, lg);\n if (!genSysP) { lg(\"WARN\", \"MetaAgent\", `LLM no content for meta-prompt tid=${tid}.`, undefined, mid, tid); return { mid, cid: tid, p: { sysPrompt: \"\" }, e: \"AI failed to generate system prompt.\" }; }\n const finalP = genSysP.replace(/^```(markdown|text)?\\s*/i, '').replace(/\\s*```$/, '').trim();\n const output: MetaOut = { sysPrompt: finalP };\n lg(\"INFO\", \"MetaAgent\", `Meta-prompt task OK tid=${tid}.`, { outputLen: output.sysPrompt.length }, mid, tid);\n return { mid, cid: tid, p: output };\n } catch (err: any) { lg(\"ERROR\", \"MetaAgent\", `Meta-prompt task fail tid=${tid}.`, { err: err.message }, mid, tid); return { mid, cid: tid, p: { sysPrompt: \"\" }, e: `Meta-prompt generation failed: ${err.message}` }; }\n}\n\nclass Orchestrator {\n baseLog: LogFn;\n mandateId: string;\n lg: LogFn;\n\n constructor(baseLogger: LogAgent) {\n this.mandateId = `m-${Date.now()}-${Math.random().toString(36).substring(2, 8)}`;\n baseLogger.setMidCtx(this.mandateId);\n this.baseLog = baseLogger.log.bind(baseLogger);\n this.lg = (lvl, comp, msg, dt?, _midIgnored?, tid?) => {\n this.baseLog(lvl, comp, msg, dt, this.mandateId, tid);\n };\n this.lg(\"INFO\", \"Orchestrator\", `Initialized mandate: ${this.mandateId}`);\n }\n\n async runWorkflow(input: UserIn): Promise {\n this.lg(\"INFO\", \"Orchestrator\", `Workflow started task: ${input.task}`, { inputKeys: Object.keys(input.data) });\n const tid = `${input.task}-${Date.now()}-${Math.random().toString(36).substring(2, 7)}`;\n const taskInput: TskIn = { mid: this.mandateId, tid: tid, p: input.data };\n let taskOutput: TskOut;\n\n try {\n const agentLog: LogFn = (lvl, comp, msg, dt?, _midOverride?, tidFromAgent?) => {\n this.lg(lvl, comp, msg, dt, this.mandateId, tidFromAgent || tid);\n };\n switch (input.task) {\n case \"create_content\": taskOutput = await runCntAgent(taskInput as TskIn, agentLog); break;\n case \"develop_strategy\": taskOutput = await runStratAgent(taskInput as TskIn, agentLog); break;\n case \"generate_logo_concept\": taskOutput = await runLogoAgent(taskInput as TskIn, agentLog); break;\n case \"build_brand_identity\": taskOutput = await runBrandAgent(taskInput as TskIn, agentLog); break;\n case \"run_simulation\": taskOutput = await runSimAgent(taskInput as TskIn, agentLog); break;\n case \"meta_prompt\": taskOutput = await runMetaAgent(taskInput as TskIn, agentLog); break;\n default: const _exCheck: never = input.task; this.lg(\"ERROR\", \"Orchestrator\", `Unsupported task type: ${_exCheck}`, { taskInput }, this.mandateId, tid); throw new Error(`Unsupported task: ${_exCheck}`);\n }\n\n if (taskOutput.e) {\n this.lg(\"WARN\", \"Orchestrator\", `Agent task ${tid} completed with error.`, { err: taskOutput.e, agentPayload: taskOutput.p });\n return { mid: this.mandateId, err: `Agent execution failed: ${taskOutput.e}`, res: taskOutput.p, dbgLog: [], details: 'Agent reported error.' };\n }\n this.lg(\"INFO\", \"Orchestrator\", `Workflow task ${tid} (${input.task}) completed successfully.`);\n return { mid: this.mandateId, res: taskOutput.p, err: undefined, dbgLog: [] };\n\n } catch (err: any) {\n this.lg(\"ERROR\", \"Orchestrator\", `Critical workflow failure task ${tid} (${input.task}).`, { err: err.message, stack: err.stack?.substring(0, 500) });\n return { mid: this.mandateId, err: `Workflow execution failed critically: ${err.message}`, details: \"Unexpected error in orchestration/agent.\", dbgLog: [] };\n }\n }\n}\n\nexport default async function(req: Request): Promise {\n const reqId = `req-${Date.now()}-${Math.random().toString(36).substring(2, 7)}`;\n const reqLogger = new LogAgent(reqId);\n const log = reqLogger.log.bind(reqLogger);\n const url = new URL(req.url);\n const commonHeaders = { \"Content-Type\": \"application/json\", \"Access-Control-Allow-Origin\": \"*\", \"Access-Control-Allow-Methods\": \"GET, POST, OPTIONS\", \"Access-Control-Allow-Headers\": \"Content-Type, Authorization\" };\n\n if (req.method === \"OPTIONS\") { log(\"INFO\", \"Handler\", \"Handling OPTIONS request\"); return new Response(null, { status: 204, headers: commonHeaders }); }\n\n try {\n if (req.method === \"GET\" && (url.pathname === \"/\" || url.pathname === \"\")) {\n log(\"INFO\", \"Handler\", \"GET / : Serving UI\");\n const html = genHtml();\n return new Response(html, { status: 200, headers: { ...commonHeaders, \"Content-Type\": \"text/html; charset=utf-8\" } });\n\n } else if (req.method === \"POST\" && url.pathname === \"/questionnaire\") {\n log(\"INFO\", \"Handler\", \"POST /questionnaire : Handling step.\");\n let reqBody; let rawReqBody = \"[Could not read body]\";\n try {\n rawReqBody = await req.text(); reqBody = JSON.parse(rawReqBody);\n if (!reqBody || typeof reqBody !== 'object' || !reqBody.task || typeof reqBody.task !== 'string') { throw new Error(\"Invalid request: 'task' (string) required.\"); }\n } catch (e: any) {\n log(\"WARN\", \"Handler\", \"Failed parse JSON body for /questionnaire\", { err: e.message, bodyAttempt: rawReqBody });\n const errLogs = reqLogger.getLogs();\n return new Response(JSON.stringify({ err: `Bad Request: Invalid JSON or missing 'task'. ${e.message}`, state: null, dbgLog: errLogs }), { status: 400, headers: commonHeaders });\n }\n const qAgent = new QAgent(log);\n const { sessionId, task, lastAnswer, state } = reqBody;\n const qResult: QResp = await qAgent.nextStep(sessionId, task, lastAnswer, state);\n qResult.state = null; // Don't send full state to client\n log(\"DEBUG\", \"Handler\", \"Questionnaire step processed.\", { sid: qResult.sid, isDone: qResult.isDone, hasErr: !!qResult.err });\n return new Response(JSON.stringify(qResult), { status: qResult.err ? 400 : 200, headers: commonHeaders });\n\n } else if (req.method === \"POST\" && url.pathname === \"/execute\") {\n log(\"INFO\", \"Handler\", \"POST /execute : Handling workflow execution.\");\n let validatedInput: UserIn; let rawPayload = \"[Could not read body]\";\n try {\n const payload = await req.json(); rawPayload = JSON.stringify(payload).substring(0, 500);\n log(\"DEBUG\", \"Handler\", \"Received payload /execute\", { payloadPreview: rawPayload });\n validatedInput = UserInSchema.parse(payload);\n log(\"INFO\", \"Handler\", \"Payload validation OK /execute.\", { task: validatedInput.task });\n } catch (error: any) {\n let errDetails = \"Invalid input format/fields.\"; let validationIssues;\n if (error instanceof z.ZodError) {\n validationIssues = error.issues; errDetails = validationIssues.map(i => `Field '${i.path.join('.')}': ${i.message}`).join('; ');\n log(\"WARN\", \"Handler\", \"Zod validation failed /execute.\", { err: errDetails, issues: validationIssues, receivedPreview: rawPayload });\n } else { errDetails = error.message; log(\"WARN\", \"Handler\", \"Error parsing/validating /execute.\", { err: errDetails, receivedPreview: rawPayload }); }\n const errLogs = reqLogger.getLogs();\n return new Response(JSON.stringify({ err: \"Invalid input data for execution.\", details: errDetails, dbgLog: errLogs, mid: `m-validate-err-${reqId}` }), { status: 400, headers: commonHeaders });\n }\n const execLogger = new LogAgent(); // Separate logger for execution\n const orchestrator = new Orchestrator(execLogger);\n const finalResult: FinalRes = await orchestrator.runWorkflow(validatedInput);\n finalResult.dbgLog = execLogger.getLogs(); // Add execution logs\n const respStatus = finalResult.err ? 500 : 200;\n log(\"INFO\", \"Handler\", `/execute workflow finished status ${respStatus}`, { mid: finalResult.mid, hasErr: !!finalResult.err });\n return new Response(JSON.stringify(finalResult), { status: respStatus, headers: commonHeaders });\n\n } else {\n log(\"WARN\", \"Handler\", `Unsupported request: ${req.method} ${url.pathname}`);\n const errLogs = reqLogger.getLogs();\n return new Response(JSON.stringify({ err: \"Not Found or Method Not Allowed\", dbgLog: errLogs }), { status: 404, headers: commonHeaders });\n }\n } catch (err: any) {\n log(\"ERROR\", \"Handler\", \"Unhandled exception during request.\", { err: err.message, stack: err.stack?.substring(0, 500) });\n const crashLogs = reqLogger.getLogs();\n return new Response(JSON.stringify({ err: \"Internal Server Error\", details: \"Unexpected server error.\", dbgLog: crashLogs, mid: `m-crash-${reqId}` }), { status: 500, headers: commonHeaders });\n }\n}","deleted","label",[98],"permalink","https://www.val.town/x/salon/svvc/code/main.tsx","extension","actionData","errors"]