Spaces:
Sleeping
Sleeping
Commit History
-
add docker file
3c2b715
-
Record converstaions and evaulate
0e7ea7d -
add ios images
34eb71a -
Add skills for the agent to use, will expand and improve as we go.
61da950 -
Remove hardcoded text features. Ai answers everything.
203dbbc -
Add generated guide images, CLI-powered responses, and image rendering
- Install @napi-rs/canvas for programmatic image generation - Create imageGenerator.js: generates 20 PNG images (keyboard diagrams + screen mockups) for all 10 guide tasks on server startup - Serve images via Express static route at /images - Wire imageUrls through WebSocket and REST responses - Render guide images in MessageBubble (keyboard + screen diagrams) - Refactor mockResponder: remove all canned text, route everything through Claude CLI, keep guide triggers for all 10 tasks - Add /images proxy to Vite config Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>0bb8699 -
Add programmatic image generator for 10 guide tasks
Adds server/core/imageGenerator.js which uses @napi-rs/canvas to generate 20 PNG instructional images (keyboard + screen mockup for each of the 10 guide tasks). Images are saved to server/assets/generated/ with an idempotency guard; generateAllGuideImages() is the main export. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>f5fafe0 -
Make the mockresponder use internal claude code account. No API Key ATM
3f4e8c5 -
Add 6 new agent tools, mock demo mode, and skill progression
New tools: suggest_next_skill, repeat_last_step, adjust_vocabulary_level, save_note_for_user, get_user_notes, restart_conversation Mock mode: when ANTHROPIC_API_KEY is not set or MOCK_MODE=true, the app returns pre-written responses that exercise visual guides, step sequences, and skill tracking — full demo without API access. Also: UserNote model, skill progression chains, Anthropic SDK tests. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>5647c24 -
Add 6 new Claude tools to agent orchestrator
Adds suggest_next_skill, repeat_last_step, adjust_vocabulary_level, save_note_for_user, get_user_notes, and restart_conversation — including tool definitions, handleFunctionCall handlers, and system prompt entries. Also imports UserNote model and skillProgression module. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>583f14c -
Update WebSocket and REST responses to include guideId and stepSequence fields
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>3a22a13 -
feat(orchestrator): add comfort adaptation, 4 new tools, classifier integration, expanded return shape
- Add taskClassifier, StepSequence, and Conversation imports - Add VALID_GUIDE_IDS constant for guide validation - Add buildComfortGuidelines() mapping comfort levels 1–5 to tailored instructions - Update buildSystemPrompt() to accept user and classification; injects comfort guidelines, task context, and urgency note - Add 4 new Gemini tool declarations: show_visual_guide, start_step_sequence, advance_step, complete_step_sequence - Update handleFunctionCall() to async with sessionId param; add handlers for all 4 new tools with guideId/stepSequence return values - Update processMessage() to run getProfileForPrompt and classifyMessage in parallel, update conversation task_type, pass user/classification to buildSystemPrompt, collect guideId/stepSequence across tool rounds - Fix getProfileForPrompt call (remove spurious second argument) - Expand return shape to { response, safetyAlert, guideId, stepSequence } Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>eeb694b -
Ai-genertaed-project-layout
f527e03 -
feat(tests): Task 6 — Jest tests for vocabularyFilter, safetyMonitor, taskClassifier
75 tests across 3 test files covering substitution levels, whole-word matching, sentence readability splitting, emergency/scam detection with DB mock, and Claude API classification with error-handling fallbacks. Added jest.config.js to target server/__tests__ and use node environment. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>53d0d6b -
Fix WebSocket security and robustness issues in server/index.js
- Add maxPayload (64KB) to WebSocketServer to prevent DoS via oversized messages - Validate msg.userId on init (must be a non-empty string) - Validate msg.text on chat (must be a non-empty string) - Guard all ws.send() calls with ws.readyState === ws.OPEN check - Add ws.on('close') and ws.on('error') handlers for connection lifecycle logging - Add process.on('SIGTERM') for graceful shutdown of WSS and HTTP server Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>b66c66f -
feat: add WebSocket server and REST routes for users and chat
Expanded server/index.js to create an HTTP server with a WebSocket server (ws) on /ws for real-time chat, using the agent orchestrator for message processing. Added server/routes/users.js (POST, GET, PUT, PUT /onboard) and server/routes/chat.js (POST REST fallback). Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>a26c1af -
Fix code review issues in agentOrchestrator (Task 3)
- Critical 1: Add MAX_TOOL_ROUNDS=10 cap on tool-use loop with error log on breach - Critical 2: Guard empty filteredResponse before Message.create to avoid crash - Important 3: Remove await from SkillEvent.create and SafetyEvent.create (sync ops) - Important 4: Remove dead User import - Important 5: Wrap entire processMessage body in outer try/catch for FALLBACK_RESPONSE on any sync error - Minor 7: Extract hardcoded model ID to CLAUDE_MODEL constant - Minor 9: Pass pre-fetched profileString into buildSystemPrompt to avoid duplicate DB query Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>25f0095 -
Task 3: Implement agent orchestrator with Claude tool-use loop
Creates server/core/agentOrchestrator.js with processMessage(text, userId) that safety-checks input, manages conversation state, calls the Claude messages API with tool-use, handles log_skill_started and flag_emergency tool calls in a loop, applies vocabulary/readability filters, and returns { response, safetyAlert }. Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>5acd2cd -
Fix Task 2 code review issues across core modules and Conversation model
- safetyMonitor: wrap SafetyEvent.create() in try/catch so safety response is always returned even on DB failure; replace overly broad "help me" keyword with "someone help me", "please help", "I need help now" to eliminate false positives on normal IT requests - taskClassifier: warn at module load when anthropicApiKey is missing - conversationState: delegate abandonStale() to Conversation.abandonStale() instead of executing raw SQL directly; validate role is 'user' or 'assistant' in addMessage() - Conversation model: add abandonStale(cutoffMinutes) method containing the stale-session query previously inlined in conversationState - userProfileManager: guard getOrCreateUser() against undefined/falsy id - vocabularyFilter: warn on unexpected vocab levels; remove dead first lastIndex assignment and its misleading comment in enforceReadability() Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>c3c7877 -
Task 2: Implement 5 core backend modules for PC Pal
Add server/core/ processing modules that the agent orchestrator will depend on: - userProfileManager.js: getOrCreateUser, updateProfile, getProfileForPrompt - vocabularyFilter.js: filterResponse (whole-word jargon substitution by vocab level), enforceReadability (20-word sentence splitting) - safetyMonitor.js: checkMessage with emergency keyword and scam pattern detection, logs to SafetyEvent - taskClassifier.js: classifyMessage via Claude API (claude-sonnet-4-20250514), returns taskType/topic/urgency - conversationState.js: session management with getOrCreateSession, closeSession, abandonStale, getSessionMessages, addMessage - server/assets/vocabulary/basicSubstitutions.json: 20 jargon-to-plain-English substitutions Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>5b7f9b4 -
Fix Task 1 code review issues: schema defaults, DB init, model validation, StepSequence JSON parsing
- Add DEFAULT (datetime('now')) to all timestamp columns in schema.sql - Require ./db/database in server/index.js so DB initializes on startup - Add NOT NULL validation guards in Conversation, Message, SkillEvent, SafetyEvent create() methods - Parse steps JSON string into array in StepSequence.findById() and findByConversationId() Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>a84d637 -
Task 1: Project setup and database schema
- Initialize Node.js/Express backend with all required dependencies (express, ws, better-sqlite3@12.8.0, @anthropic-ai/sdk, uuid, dotenv, cors) and dev dependencies (jest, nodemon, concurrently) - Scaffold React + Vite frontend in client/ with proxy config for /api and /ws pointing to localhost:3001 - Create server/db/schema.sql with all 6 tables: users, conversations, messages, step_sequences, skill_events, safety_events - Create server/db/database.js with better-sqlite3 initialization, WAL mode, FK enforcement, and schema auto-execution - Create server/config.js for centralized env config via dotenv - Create all 6 model files (User, Conversation, Message, SkillEvent, SafetyEvent, StepSequence) with full CRUD operations - Add .env.example and .gitignore Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>0737291