Christian Gick e04a8ab524 feat(CF-580): Implement transcript-based session recovery for MCP ...

Add direct transcript ingestion and orphan recovery using Claude Code's JSONL
transcripts instead of relying on daemon-based note synchronization.

Changes:

1. **Database migration** (027_session_transcript_storage.sql):
   - Add transcript_jsonl, transcript_ingested_at, transcript_file_path columns
   - Add indexes for efficient ingestion tracking

2. **Transcript parser utility** (src/utils/transcript-parser.ts):
   - parseTranscriptFile(): Parse JSONL line-by-line, handle corrupt lines
   - encodeWorkingDir(): Convert paths to Claude Code directory encoding
   - findOrphanedTranscripts(): Scan for stale transcript files
   - ingestTranscriptToDatabase(): Main ingestion function for Node.js

3. **Orphan recovery enhancement** (src/tools/sessions.ts):
   - sessionRecoverOrphaned() now tries transcript ingestion first
   - Finds most recently modified JSONL in project directory
   - Falls back to legacy notes.md recovery for backward compatibility
   - Properly handles path encoding (/ and . → -)

Benefits:
- No daemon needed for recovery (Phase 2 will remove LaunchAgent)
- Full transcript audit trail stored in database
- Immediate recovery capability for orphaned sessions
- Cleaner architecture (no markdown parsing complexity)
- Compatible with Claude Code's UUID-based session files

Co-Authored-By: Claude Haiku 4.5 <noreply@anthropic.com>

2026-01-29 17:53:37 +02:00

1.2 KiB

Raw Permalink Blame History

View Raw