Move webhook tests to root tests/ dir and restructure to be non-flaky. Setup webhook capture server once via preload, tear down on process exit. Tests now clean up their own PRs in afterEach. Added bridge test to verify Spike PRs create Discord threads. Made Discord client login explicit via startDiscord(). Updated config with new dev Discord channel and Spike username mappings.
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
Implement a centralized logging system with typed events (webhook, PR, comment, review, Discord relay, errors, etc.) that feeds a pub/sub pattern. Events are logged to console (with ANSI colors: green=normal, yellow=skipped, red=errors, cyan=PR#) and persisted to JSONL files organized by git commit SHA.
Restructure server code into src/server/ directory with separate route handlers. Add a new /logs endpoint with an HTML viewer that displays events in a searchable table with type/repo filtering and browser timezone conversion. Remove the old /errors endpoint.
Improvements:
- All webhook handlers, Discord events, and errors now use log() instead of console.* calls
- Error objects are properly serialized to JSONL (message + stack extracted)
- Path traversal vulnerability fixed in readLogFile()
- Failed API calls and caught errors are now always logged
- Graceful error handling with fallbacks where appropriate (e.g., missing Discord users)
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
Split the monolithic gitea/helpers.ts (which had Discord imports and created circular dependencies) into three focused libs:
1. **gitea/** — Pure API client: fetchPR, fetchReviewComments, convertUsername, threadName. No side effects or external deps.
2. **discord/** — Discord client setup: bot login, event listeners, slash commands. Now isolated from Gitea internals.
3. **bridge/** — New integration layer: webhook handler, DB mappings (Gitea PR ↔ Discord thread), Discord helpers, and createPRComment.
Dependencies now flow one direction: bridge → gitea and bridge → discord. No circular imports.
Added:
- Barrel exports (index.ts) for each lib with public API
- README.md for each lib documenting the barrel exports
- Comprehensive spike README.md with setup guide and architecture explanation
- Integration tests for webhooks (callback-based, no race conditions)
- Unit tests for pure API functions
- CLAUDE.md with links to each lib's README
This architecture makes it possible for AI to understand a lib by reading just its README, keeping context focused and small.
Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>
