Launch checklist implementation:
- Add Privacy Policy page (/privacy) with GDPR compliance
- Add Terms of Service page (/terms)
- Add custom 404 and error pages with helpful navigation
- Add FAQ page (/faq) covering common questions
- Add SEO meta tags with OpenGraph/Twitter cards
- Add JSON-LD structured data (Organization, WebSite, SoftwareApplication)
- Add sitemap.ts and robots.ts for search engines
- Add security headers (HSTS, CSP, X-Frame-Options) in vercel.json
- Add security.txt at /.well-known/security.txt
- Add API rate limiting (100 req/min default, 20 req/min strict)
- Add empty states in tool search for better UX
- Update AppHeader with FAQ link
- Update AppFooter with Privacy/Terms links
- Update biome.json to allow dangerouslySetInnerHTML for JSON-LD in page files
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
The executor returns HTTP 500 for all tool errors, including missing env vars.
Before: 500 response = BROKEN
After: Check error message for env/validation patterns before marking BROKEN
This ensures tools that require API keys (like @parallel-web/ai-sdk-tools)
show importHealth: HEALTHY since the tool loads correctly - it just needs config.
Simplify execution health check logic:
- If executor responds (2xx or 4xx), tool is HEALTHY
- Only mark BROKEN for 5xx errors or network/timeout failures
- Validation errors (URL format, missing fields) mean tool IS working
- Remove brittle pattern matching for specific error messages
The previous approach tried to match specific error patterns to determine
if an error was "acceptable". This was fragile. The new approach:
- Import check: can we load and describe the tool?
- Execution check: did the tool execute at all?
If a tool throws a validation error, it executed successfully - it's
correctly rejecting invalid test input. Only infrastructure failures
(executor down, network timeout) indicate a truly broken tool.
Tools that reject invalid test inputs (like invalid URLs) are actually
working correctly - they're validating input as expected. Previously
the health check marked these as BROKEN because the test used dummy
values like 'test' which failed Zod validation.
Now input validation errors (invalid URL, invalid format, type mismatch,
etc.) are treated as HEALTHY, similar to how we already treat missing
environment variables.
Tools that fail due to missing environment variables (API keys, etc.)
are not actually broken - they just need configuration. Added detection
for common env var error patterns and mark these tools as HEALTHY
instead of BROKEN.
Error patterns detected:
- 'is required'
- 'is not set'
- 'missing environment'
- 'API key required/not provided'
- etc.
This fixes false positives where tools like @superagent-ai/ai-sdk
were marked as broken when they just need SUPERAGENT_API_KEY configured.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- Create ToolHealthBadge component in @tpmjs/ui for broken tool indicator
- Create ToolHealthBanner component in @tpmjs/ui for detailed health warnings
- Integrate both components into playground ToolsSidebar:
- Badge shows in tool cards in left sidebar
- Banner shows in tool detail modal
- Update search-registry to include health fields in API responses
- Add package.json exports for new health components
These components provide consistent UI for displaying broken tool status
across the application (tool search, tool detail pages, playground).
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Add comprehensive health monitoring for TPMJS tools that tracks both
import and execution health via Railway executor service.
## Database Schema
- Add HealthStatus enum (UNKNOWN, HEALTHY, BROKEN)
- Add HealthCheckType enum (IMPORT, EXECUTION, FULL)
- Add health fields to Tool model:
- importHealth: tracks if tool can be loaded
- executionHealth: tracks if tool can execute
- lastHealthCheck: timestamp of last check
- healthCheckError: stores error message
- Add HealthCheck audit table for full history
## Core Service
Create health-check-service.ts with 5 functions:
1. checkImportHealth() - Tests tool loading via /load-and-describe
2. checkExecutionHealth() - Tests execution via /execute-tool
3. generateTestParameters() - Creates minimal test params by type
4. performHealthCheck() - Full check with database updates
5. performBatchHealthCheck() - Processes tools in batches
Features:
- 30-second timeout per check
- Skips execution if import fails
- Batch processing (5 concurrent, 1s delays)
- Full audit trail in HealthCheck table
## API Endpoints
/api/sync/health-check (POST):
- Daily cron job at 2am UTC
- Checks all tools in database
- Requires CRON_SECRET auth
- Logs results to SyncLog table
- Max duration: 5 minutes
/api/tools/broken (GET):
- Lists all tools with broken health status
- Filters by importHealth='BROKEN' OR executionHealth='BROKEN'
- Includes package metadata
- Orders by lastHealthCheck DESC
## Configuration
- Add RAILWAY_EXECUTOR_URL to env.ts
- Add daily cron job to vercel.json
- Use db:push for schema changes (existing production data)
Next: Manual trigger endpoint + health filtering + UI components
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
BREAKING CHANGE: Complete refactoring from single-tool to multi-tool package support
Database Schema:
- Split Tool model into Package (1) and Tool (many) with one-to-many relationship
- Package stores npm metadata and package-level tpmjs fields (category, env, frameworks, tier)
- Tool stores individual tool exports with tool-level metadata (exportName, description, parameters, returns, aiAgent)
- Unique constraint on (packageId, exportName) to prevent duplicate tools
- Cascade deletes when packages are removed
Type System:
- Updated tpmjs field schema to support tools array
- Each tool has exportName, description, parameters, returns, aiAgent
- Package-level fields: category, env, frameworks shared across all tools
- Backward compatible with legacy single-tool format (auto-migrates to exportName: "default")
API Updates:
- Updated all /api/tools routes to query Tool model with Package relations
- Updated /api/tools/[slug] to accept package/export path segments
- Updated tool-executor-agent to use actual exportName instead of hardcoded "default"
- Updated metrics sync to calculate quality scores per Tool
Frontend Updates:
- Updated tool search page to display exportName as primary heading
- Updated tool detail pages to show package name as secondary info
- Removed tag-based filtering (tags moved to package level)
Manual Tool Registry:
- Added manual-tools.ts with 23 curated tools from major providers
- Created sync-manual-tools.ts script to sync manual tools to database
- Added MANUAL_TOOLS.md documentation for manual tool system
- Added GitHub workflow for automated daily sync
- Includes tools from: Vercel, Exa, Firecrawl, AWS Bedrock, Perplexity, Tavily, Superagent, Valyu
Playground Updates:
- Updated tool loader to load multiple tools per package
- Added sanitizeToolName for OpenAI API compatibility
Sync System Updates:
- Updated changes feed sync to handle multi-tool packages
- Updated keyword sync to upsert multiple tools per package
- Added orphaned tool deletion when tools removed from package.json
Migration Strategy:
- Database uses same Neon instance for dev and prod
- Schema updated via prisma db push (no migration files yet)
- All data repopulates from npm via sync system
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Major changes:
- Fixed package executor URL protocol handling (add https:// if missing)
- Switched from streamText to generateText for proper tool execution
- AI now calls tool AND generates natural language response
- Tool results no longer show raw JSON metadata
Tool executor (tool-executor-agent.ts):
- Use generateText() instead of streamText() for tool execution
- Add system prompt to guide AI to summarize tool results
- Return result.text for natural language output
- Tool definition uses inputSchema (AI SDK v6 format)
Package executor:
- Add getSandboxUrl() to ensure URL has https:// protocol
- Fixes "Failed to parse URL" error in production
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
- In AI SDK v6, tool results are in fullResponse.messages with role 'tool'
- Updated result extraction to iterate through messages array
- Added detailed logging to debug response structure
- Handle both text output and tool-only responses
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
OpenAI was rejecting tool definitions with error "schema must be a JSON Schema of 'type: "object"'". This was caused by AI SDK v5 not properly converting Zod schemas to JSON Schema format.
**Changes:**
- Upgrade AI SDK from v5.0.104 to v6.0.0-beta.124
- Upgrade @ai-sdk/openai from v2.0.74 to v3.0.0-beta.22
- Upgrade Zod from v3.25.76 to v4.1.13 across all packages
**AI SDK v6 breaking changes:**
- Tool definition API: `parameters` renamed to `inputSchema`
- Removed `aiTool()` wrapper - use plain object with description, inputSchema, execute
- Streaming API: Use `textStream` async iterator instead of onChunk callback
- Zod schemas now properly converted to JSON Schema for OpenAI
**Zod v4 breaking changes:**
- `z.record()` now requires two arguments: `z.record(keySchema, valueSchema)`
- `z.enum()` params changed: `errorMap` removed, use `message` instead
- Type system improvements require explicit type parameters
- Fixed type errors in @tpmjs/env, @tpmjs/npm-client, @tpmjs/types
**Files changed:**
- apps/web/src/lib/ai-agent/tool-executor-agent.ts
- Updated tool definition to use `inputSchema` instead of `parameters`
- Removed `aiTool()` wrapper
- Fixed streaming to use `textStream` iterator
- packages/env/src/index.ts
- Updated type constraint from `z.ZodRawShape` to `Record<string, z.ZodTypeAny>`
- packages/npm-client/src/package.ts
- Fixed `z.record()` calls to include both key and value schemas
- Added type assertions for record indexing
- packages/types/src/tpmjs.ts
- Changed `errorMap` to `message` in z.enum() calls
**Testing:**
- ✅ Type-check passes
- ✅ Production build succeeds
- ✅ All routes compile correctly
This fixes the tool execution error where OpenAI rejected tool schemas with invalid format.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Add console.log statements to track:
- Tool parameters array and length
- Generated Zod schema details
- Tool definition structure
- Sanitized tool name
- Complete tools config sent to OpenAI
This will help diagnose why OpenAI is still receiving 'type: "None"'
for empty parameter schemas despite the fix.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Problem: OpenAI API error 'got type: None' when tool has no/invalid parameters
Solution: Add guard to explicitly create empty object schema with description when no parameters exist
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
Problem: OpenAI tool names must match ^[a-zA-Z0-9_-]+ but npm package names like @tpmjs/createblogpost contain @ and /
Solution: Add sanitizeToolName() function and update pnpm-lock.yaml
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
**Problem:**
The Interactive Playground was failing with "Missing tiktoken_bg.wasm" error in production. Tiktoken requires WASM files which don't work in Vercel's serverless environment.
**Solution:**
- Remove tiktoken import from tool-executor-agent
- Replace tiktoken-based token counting with character estimation (~4 chars/token)
- Remove tiktoken from package.json dependencies
- Remove unused biome-ignore comment
**Impact:**
- Token counting is now approximate but consistent
- No more WASM-related runtime errors
- Serverless deployment works properly
- Tool execution now functional in production
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
**Problem:**
The Interactive Playground was failing with "Missing tiktoken_bg.wasm" error in production. Tiktoken requires WASM files which don't work in Vercel's serverless environment.
**Solution:**
- Remove tiktoken import from tool-executor-agent
- Replace tiktoken-based token counting with character estimation (~4 chars/token)
- Remove tiktoken from package.json dependencies
- Remove experimental webpack WASM config (no longer needed)
**Impact:**
- Token counting is now approximate but consistent
- No more WASM-related runtime errors
- Serverless deployment works properly
- Tool execution now functional in production
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
