dellis317 · dellis317 · Feb 28, 2026 · Feb 28, 2026 · Feb 28, 2026 · Feb 28, 2026
diff --git a/.gitignore b/.gitignore
@@ -1,6 +1,8 @@
 node_modules
 dist
+data/
 .DS_Store
 server/public
 vite.config.ts.*
-*.tar.gz
+*.tar.gz
+.env
diff --git a/.mcp.json b/.mcp.json
@@ -0,0 +1,15 @@
+{
+  "mcpServers": {
+    "aiqa": {
+      "command": "npx",
+      "args": [
+        "aiqastudio-mcp"
+      ],
+      "env": {
+        "AIQA_API_KEY": "aiq_425865884d0ef800ba96bab750fc43f6",
+        "AIQA_PROJECT_ID": "eacc71c9-f5f5-4d1b-a798-3f47b99a04da",
+        "AIQA_URL": "https://aiqastudio.com"
+      }
+    }
+  }
+}
diff --git a/CLAUDE.md b/CLAUDE.md
diff --git a/UNUSED_APIS.md b/UNUSED_APIS.md
@@ -0,0 +1,122 @@
+# Unused API Endpoints
+
+**Generated**: 2026-02-28
+**Method**: Cross-referenced all route definitions in `server/routes.ts` against all API calls in `client/src/`
+
+---
+
+## Summary
+
+| Category | Count |
+|----------|-------|
+| Persona query/management APIs with no UI | 6 |
+| Persona admin APIs with no UI | 2 |
+| Agent CRUD APIs with no UI | 5 |
+| Non-streaming agent execute | 1 |
+| Admin analytics/config with no UI | 6 |
+| Misc unused | 3 |
+| **Total unused from UI** | **23** |
+| External-only (webhooks, health checks) | 3 |
+| Dead code (replit_integrations never imported) | 11 routes across 3 files |
+
+---
+
+## Unused Endpoints — No Frontend Callers
+
+### Persona Query & Management (6 endpoints)
+
+| Method | Endpoint | Line | Notes |
+|--------|----------|------|-------|
+| GET | `/api/personas` | routes.ts:937 | Base personas list — UI imports from `shared/personas.ts` directly |
+| GET | `/api/personas/hierarchy` | routes.ts:950 | Hierarchy view — no UI |
+| GET | `/api/personas/all` | routes.ts:969 | All personas including DB overrides — no UI |
+| GET | `/api/personas/domain/:domain` | routes.ts:979 | Filter by domain — no UI |
+| GET | `/api/personas/:personaId/versions` | routes.ts:1011 | Version history — no UI |
+| PUT | `/api/admin/personas/:personaId` | routes.ts:1045 | Full persona update — Admin.tsx only uses PATCH lock, not PUT |
+
+### Persona Admin (2 endpoints)
+
+| Method | Endpoint | Line | Notes |
+|--------|----------|------|-------|
+| DELETE | `/api/admin/personas/:personaId/override` | routes.ts:1127 | Delete persona override — no UI |
+| GET | `/api/admin/personas/export` | routes.ts:1141 | Export all personas as JSON — no UI |
+
+### Agent CRUD (5 endpoints)
+
+The frontend only uses `POST /api/agents/:agentId/execute/stream` and `POST /api/agents/execute-inline`. All CRUD management endpoints are unused.
+
+| Method | Endpoint | Line | Notes |
+|--------|----------|------|-------|
+| POST | `/api/agents` | routes.ts:1157 | Create agent — no UI |
+| GET | `/api/agents` | routes.ts:1182 | List agents — no UI |
+| GET | `/api/agents/:agentId` | routes.ts:1201 | Get single agent — no UI |
+| PUT | `/api/agents/:agentId` | routes.ts:1218 | Update agent — no UI |
+| DELETE | `/api/agents/:agentId` | routes.ts:1242 | Delete agent — no UI |
+
+### Agent Execution (1 endpoint)
+
+| Method | Endpoint | Line | Notes |
+|--------|----------|------|-------|
+| POST | `/api/agents/:agentId/execute` | routes.ts:1262 | Non-streaming execute — UI only uses `/execute/stream` |
+
+### Admin Analytics & Config (6 endpoints)
+
+| Method | Endpoint | Line | Notes |
+|--------|----------|------|-------|
+| GET | `/api/admin/agent-overrides` | routes.ts:1395 | List all agent overrides — UI fetches via `agent-prompts` instead, then does individual PUT/PATCH/DELETE on `:taskType` |
+| GET | `/api/admin/persona-usage` | routes.ts:1674 | Persona usage stats — no UI |
+| GET | `/api/admin/event-breakdown` | routes.ts:1688 | Event breakdown — no UI |
+| GET | `/api/admin/llm-logs` | routes.ts:1717 | LLM call logs (admin) — no UI |
+| GET | `/api/admin/voice-capture-config` | routes.ts:1886 | Admin voice config getter — no UI (user `GET /api/voice-capture-config` IS used) |
+| PUT | `/api/admin/voice-capture-config` | routes.ts:1897 | Admin voice config setter — no UI |
+
+### Misc Unused (3 endpoints)
+
+| Method | Endpoint | Line | Notes |
+|--------|----------|------|-------|
+| GET | `/api/llm-status` | routes.ts:607 | Health/diagnostic endpoint — no frontend caller |
+| GET | `/api/llm-logs` | routes.ts:1740 | Non-admin LLM logs — no UI |
+| POST | `/api/tracking/events` | routes.ts:1536 | Batch event tracking — UI uses singular `POST /api/tracking/event` |
+
+---
+
+## External-Only Endpoints (NOT dead code)
+
+These are called by external services or server-side processes, not the frontend:
+
+| Method | Endpoint | Line | Called By |
+|--------|----------|------|-----------|
+| POST | `/api/stripe/webhook` | routes.ts:5044 | Stripe webhook callbacks |
+| GET | `/api/stripe/payments` | routes.ts:5090 | Admin/external tooling (no frontend caller) |
+| POST | `/api/chat` | routes.ts:3586 | Non-streaming chat — possibly for external integrations (UI uses `/api/chat/stream`) |
+
+### Server-Side Auto-Sync (called on server startup, not from UI)
+
+| Method | Endpoint | Line | Notes |
+|--------|----------|------|-------|
+| POST | `/api/admin/sync-app-docs` | routes.ts:1973 | Auto-syncs on server startup for admin user |
+| POST | `/api/admin/sync-persona-docs` | routes.ts:2130 | Manual admin trigger |
+
+---
+
+## Dead Code: Replit Integrations (never imported)
+
+These route-registration functions exist in `server/replit_integrations/` but are **never imported or called** from `server/index.ts` or `server/routes.ts`:
+
+| File | Function | Routes Defined |
+|------|----------|---------------|
+| `server/replit_integrations/audio/routes.ts` | `registerAudioRoutes` | 5 conversation/audio endpoints |
+| `server/replit_integrations/chat/routes.ts` | `registerChatRoutes` | 5 conversation/chat endpoints |
+| `server/replit_integrations/image/routes.ts` | `registerImageRoutes` | 1 image generation endpoint |
+
+These files can be safely removed.
+
+---
+
+## Recommendations
+
+1. **Remove dead code**: The `server/replit_integrations/` routes are never registered — safe to delete
+2. **Remove or build UI**: The 23 unused endpoints fall into two camps:
+   - **Build admin panels**: Persona management, agent CRUD, analytics (llm-logs, persona-usage, event-breakdown) could power a richer admin dashboard
+   - **Remove safely**: Non-streaming `POST /api/agents/:agentId/execute`, batch `POST /api/tracking/events`, `GET /api/llm-status` can likely be removed
+3. **Document external endpoints**: `stripe/webhook`, `stripe/payments`, `sync-app-docs`, `sync-persona-docs`, and non-streaming `POST /api/chat` should be documented as external/server-only APIs
diff --git a/apps/agent-editor/CLAUDE.md b/apps/agent-editor/CLAUDE.md
@@ -0,0 +1,96 @@
+# Agent Editor — App Guide
+
+> **Template ID**: `agent-editor` | **Category**: `build` | **Layout**: `standard`
+
+## Purpose
+
+Design **multi-step AI agent workflows** with structured steps (Input → Actor → Output). Each step is a prompt execution unit with defined inputs, processing logic, and outputs that chain together. Includes a built-in execution engine for testing workflows with real LLM calls, token monitoring, and SSE-based streaming results.
+
+## User Workflow
+
+1. **Select** the Agent Editor template from the landing page
+2. **Define the agent** — What should this agent do? First step input? Final output? How many steps?
+3. **Auto-interview starts** with `architect` persona (not `thinking_bigger`)
+4. **Steps tab opens first** — Build step sequence with StepBuilder component
+5. **Configure each step** — Define Input, Actor (prompt), Output for each step
+6. **Test** — Execute the workflow via AgentRunner with real LLM calls
+7. **Monitor** — Watch step-by-step progress via SSE streaming, track token usage
+8. **Provoke & refine** — Personas challenge prompt quality, step compatibility, token budget
+9. **Output** — A complete agent workflow definition stored in the database
+
+## Three-Layer Definition
+
+### Layer 1: PrebuiltTemplate (`client/src/lib/prebuiltTemplates.ts`)
+- **Title**: Agent Editor
+- **Short Label**: Agent
+- **Subtitle**: Design multi-step AI workflows
+- **Objective**: "Design a multi-step AI agent workflow with clear input, processing, and output for each step"
+- **Draft Questions**:
+  1. What should this agent accomplish end-to-end?
+  2. What is the first step's input?
+  3. What is the final output?
+  4. How many steps do you envision?
+- **Steps**: `[define, steps, test]` (3 steps)
+- **Provocation Sources**: Workflow Architect, Prompt Quality Reviewer
+- **Template Content**: Empty (workflows built through step builder)
+
+### Layer 2: AppFlowConfig (`client/src/lib/appWorkspaceConfig.ts`)
+- **Workspace Layout**: `standard`
+- **Default Toolbox Tab**: **`steps`** — unique tab, opens first
+- **Auto-Start Interview**: `true`
+- **Auto-Start Personas**: **`["architect"]`** — uses Architect persona, not thinking_bigger
+- **Left Panel Tabs**: `[steps, provoke, context]` — **Steps tab first** (unique)
+- **Right Panel Tabs**: `[execution, discussion]` — **Execution tab** (unique)
+- **Writer Config**:
+  - Mode: `edit`
+  - Output Format: `markdown`
+  - Document Type: "agent workflow definition"
+  - Feedback Tone: "precise and architecture-focused"
+
+### Layer 3: Context Builder (`server/context-builder.ts`)
+- **Document Type**: "agent workflow definition"
+- **Feedback Tone**: "precise and architecture-focused"
+- **Output Format**: `markdown`
+- **System Guidance**: Multi-step agent design with Input → Actor → Output pattern. Challenges vague prompts (what exactly does "analyze" mean?), flags token budget issues (will this step exceed limits?), ensures step-to-step compatibility (does step 2's input match step 1's output?), and validates the entire chain end-to-end.
+
+## App-Specific Components
+
+| Component | Path | Purpose |
+|-----------|------|---------|
+| `StepBuilder.tsx` | `client/src/components/StepBuilder.tsx` | Left panel "Steps" tab — build, reorder, configure agent steps with token counter |
+| `AgentRunner.tsx` | `client/src/components/AgentRunner.tsx` | Right panel "Execution" tab — input field, run button, SSE-based step-by-step progress, results display |
+| `TokenCounter.tsx` | `client/src/components/TokenCounter.tsx` | Token estimation and usage display |
+
+## App-Specific API Endpoints
+
+| Endpoint | Method | Purpose |
+|----------|--------|---------|
+| `/api/agents` | POST | Create agent definition |
+| `/api/agents` | GET | List user's agents |
+| `/api/agents/:agentId` | GET | Get single agent |
+| `/api/agents/:agentId` | PUT | Update agent |
+| `/api/agents/:agentId` | DELETE | Delete agent |
+| `/api/agents/:agentId/execute` | POST | Execute saved agent (non-streaming) |
+| `/api/agents/:agentId/execute/stream` | POST | Execute saved agent (SSE streaming) |
+| `/api/agents/execute-inline` | POST | Execute unsaved agent definition (for testing) |
+
+**Server-side engine**: `server/agent-executor.ts` — Orchestrates step-by-step execution with LLM calls, handles streaming, token tracking, and error recovery.
+
+## Data Model
+
+Agents are stored as database entities (not just documents):
+- **Agent definition**: Name, description, steps array, userId
+- **Each step**: stepId, label, input template, actor prompt, output schema
+- **Execution history**: Input, output, token usage, duration per step
+
+## Key Behaviors
+
+- **Architect persona**: Only app that auto-starts with `architect` instead of `thinking_bigger`
+- **Steps tab first**: Only app with a "Steps" panel in the left toolbox
+- **Execution tab**: Only app with an "Execution" panel in the right side
+- **Database-backed**: Agent definitions are persisted in the database (CRUD), not just document content
+- **Real execution**: Can actually run the designed workflow with real LLM calls
+- **SSE streaming**: Execution results stream step-by-step via Server-Sent Events
+- **Token monitoring**: Tracks and displays token usage per step and total
+- **Chain validation**: System guidance specifically validates Input → Output compatibility between adjacent steps
+- **Inline execution**: Can test unsaved agent definitions without persisting first
diff --git a/apps/bs-chart/CLAUDE.md b/apps/bs-chart/CLAUDE.md
@@ -0,0 +1,58 @@
+# BS Chart — Infinite Canvas Diagramming Tool
+
+## Identity
+
+**Purpose**: An infinite canvas diagramming tool for creating data flow diagrams, architecture charts, ERD schemas, and process maps.
+
+**Category**: build
+
+**Philosophy**: Visual thinking should be as natural as whiteboarding — drag shapes, connect them, annotate with text. Voice commands let you design hands-free.
+
+**User Workflow**: User drags shapes from the left toolbar onto the canvas, connects them with click-and-drag connectors, edits properties in the right panel, and can issue voice commands to manipulate the chart.
+
+## Three-Layer Definition
+
+### Layer 1: Template (`prebuiltTemplates.ts`)
+- **id**: `bs-chart`
+- **category**: `build`
+- **icon**: `GitGraph`
+- **statusLabel**: `alpha`
+
+### Layer 2: Workspace Config (`appWorkspaceConfig.ts`)
+- **workspaceLayout**: `bs-chart` (custom layout — toolbar | canvas | properties)
+- **defaultToolboxTab**: `context`
+- **autoStartInterview**: false
+- **writer.mode**: `edit`
+- **writer.outputFormat**: `markdown`
+
+### Layer 3: Context Builder (`context-builder.ts`)
+- **documentType**: `diagram / chart`
+- **feedbackTone**: `precise and architecture-focused`
+- Challenges focus on: completeness, labeled connectors, orphaned nodes, table schema validity
+
+## App-Specific Components
+
+| Component | File | Purpose |
+|-----------|------|---------|
+| `BSChartWorkspace` | `bschart/BSChartWorkspace.tsx` | Top-level workspace: toolbar, canvas, properties |
+| `BSChartCanvas` | `bschart/BSChartCanvas.tsx` | Infinite canvas with grid, pan, zoom, node/connector rendering |
+| `BSChartToolbar` | `bschart/BSChartToolbar.tsx` | Left panel: tools, shapes, actions |
+| `BSChartProperties` | `bschart/BSChartProperties.tsx` | Right panel: edit selected node/connector |
+| `BSNodeRenderer` | `bschart/nodes/BSNodeRenderer.tsx` | Renders all node types (table, diamond, rect, text, badge) |
+| `BSConnectorLayer` | `bschart/BSConnectorLayer.tsx` | SVG overlay rendering connectors with markers |
+| `useChartState` | `bschart/hooks/useChartState.ts` | Chart state management with undo/redo |
+| `useCanvasInteraction` | `bschart/hooks/useCanvasInteraction.ts` | Pan, zoom, drag, marquee, connect interactions |
+| `useVoiceChartCommands` | `bschart/hooks/useVoiceChartCommands.ts` | Voice command parser and executor |
+| `types.ts` | `bschart/types.ts` | Data model: BSNode, BSConnector, BSChart |
+
+## Key Behaviors
+
+- **Infinite Canvas**: Pan (middle-click/hand tool), zoom (scroll wheel), grid snapping
+- **Shape Types**: table, diamond, rectangle, rounded-rect, text, badge
+- **Connectors**: Orthogonal bezier paths with configurable arrow markers (arrow, diamond, circle)
+- **Undo/Redo**: Full history stack with Ctrl+Z/Ctrl+Y
+- **Keyboard Shortcuts**: V=select, H=pan, C=connect, R=rect, D=diamond, T=table, etc.
+- **Voice Commands**: "add table called Users", "connect Users to Orders", "move Orders right of Users"
+- **Table Nodes**: Header row, dynamic columns/rows, per-cell editing, striped rows
+- **Properties Panel**: Full style editing (fill, stroke, text color, font, opacity, border radius)
+- **Export/Import**: JSON chart files, save to Context Store for sharing
diff --git a/apps/email-composer/CLAUDE.md b/apps/email-composer/CLAUDE.md
@@ -0,0 +1,73 @@
+# Email Composer — App Guide
+
+> **Template ID**: `email-composer` | **Category**: `write` | **Layout**: `standard`
+
+## Purpose
+
+Compose **polished business emails** quickly. Tone adapts to the described recipient (C-suite, peers, clients, direct reports). Uses a unique `email` output format (not markdown) that includes Subject line, greeting, body, and sign-off. No template — starts freeform. Challenges focus on clarity, tone appropriateness, and actionability.
+
+## User Workflow
+
+1. **Select** the Email Composer template from the landing page
+2. **Define the email** — Purpose? Recipient? Key message or ask?
+3. **Write** — Compose directly in the freeform editor
+4. **Provoke tab** generates challenges against tone, clarity, and actionability
+5. **Iterate** — refine tone, tighten the ask, improve structure
+6. **Output** — A professional email with Subject, greeting, body, sign-off
+
+## Three-Layer Definition
+
+### Layer 1: PrebuiltTemplate (`client/src/lib/prebuiltTemplates.ts`)
+- **Title**: Email Composer
+- **Short Label**: Email
+- **Subtitle**: Business professional emails, fast
+- **Objective**: "Compose a clear, professional business email that achieves the stated communication goal with appropriate tone for the audience"
+- **Draft Questions**:
+  1. What is the purpose of this email?
+  2. Who is the recipient (role, relationship)?
+  3. What is the key message or ask?
+- **Steps**: `[{ id: "compose", label: "Compose your email" }]`
+- **Provocation Sources**: Communications Director, Executive Assistant, Recipient's Perspective
+- **Template Content**: Empty (freeform email)
+
+### Layer 2: AppFlowConfig (`client/src/lib/appWorkspaceConfig.ts`)
+- **Workspace Layout**: `standard`
+- **Default Toolbox Tab**: `provoke`
+- **Auto-Start Interview**: `false`
+- **Auto-Start Personas**: none
+- **Left Panel Tabs**: `[provoke, context]`
+- **Right Panel Tabs**: `[discussion]`
+- **Writer Config**:
+  - Mode: `edit`
+  - Output Format: **`email`** — unique format (not `markdown`)
+  - Document Type: "business email"
+  - Feedback Tone: "direct and professional — focus on clarity, tone, and actionability"
+
+### Layer 3: Context Builder (`server/context-builder.ts`)
+- **Document Type**: "business email"
+- **Feedback Tone**: "direct and professional — focus on clarity, tone, and actionability"
+- **Output Format**: **`email`**
+- **System Guidance**: Formats output as email with Subject line, greeting, body, sign-off. Adapts tone based on recipient context:
+  - C-suite: concise, data-driven, decision-focused
+  - Peers: collaborative, direct
+  - Clients: professional, service-oriented
+  - Direct reports: clear, supportive, actionable
+  - Front-loads purpose in the first sentence. Forbids generic fillers ("I hope this email finds you well").
+
+## Components Used
+
+All shared/standard components — no app-specific components.
+
+## API Endpoints Used
+
+All shared endpoints only.
+
+## Key Behaviors
+
+- **Unique output format**: Only app that uses `email` format instead of `markdown` — output includes Subject line structure
+- **No auto-interview**: Emails are short — no interview warmup needed
+- **No template**: Starts blank — user composes from scratch
+- **Recipient-adaptive tone**: System guidance adjusts formality based on described recipient
+- **Anti-filler enforcement**: System guidance explicitly forbids generic email openers
+- **Action-focused**: Every email must have a clear ask or next step
+- **Minimal provocation sources**: Only 3 (vs. 5 for most apps) — emails are focused documents