feat: memory extraction pipeline (#261)
* feat: add memory extraction pipeline extension
Two-phase pipeline that extracts durable knowledge from session transcripts
and consolidates into project-scoped memory artifacts injected into future
sessions via system prompt.
- MemorySettings in settings-manager (disabled by default)
- SQLite storage with lease-based job queue (better-sqlite3)
- Phase 1: scan .jsonl sessions, extract knowledge via LLM
- Phase 2: consolidate extractions into MEMORY.md
- /memory command: view, clear, rebuild, stats
- Secret redaction on all stored output
- Watermark tracking to skip unchanged sessions
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
* fix: harden memory extraction pipeline security and performance
- Expand secret redaction patterns (Stripe, JWT, PEM, npm, Anthropic, OpenAI keys)
- Stream-read session file headers instead of loading entire file for cwd check
- Add 50MB file size cap to prevent OOM with concurrent extraction workers
- Delete orphaned prompt .md files (prompts are inlined in pipeline.ts)
- Reset package-lock.json to current main to fix version drift
Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
---------
Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
8ac5a82409