| name | swarm-transcript-review |
| description | Alias for swarm-transcript-scan — audits a long chat transcript for missed/unfinished action items. Use when the user types /swarm-transcript-review or /swarm-transcript-scan. |
/swarm-transcript-review — alias of /swarm-transcript-scan
This is a thin alias. The canonical skill is swarm-transcript-scan —
both names do the same thing: chunk a session JSONL transcript and run each
chunk through the swarm LLM to enumerate every action item with a
DONE / OPEN / UNCLEAR status, so a request made mid-session is not silently
dropped.
What to do
Invoke the swarm-transcript-scan skill and follow its steps exactly. In
short:
- Pick the transcript — if the user gave a path use it, else default to the
most recent session JSONL:
ls -t "$HOME/.claude/projects/"*/*.jsonl 2>/dev/null | head -5
- Run the scanner from the repo root:
python tools/swarm/transcript_action_scan.py <transcript.jsonl> \
--out reports/transcript_scan_<id>.md
- Report the
## OPEN (...) section + chunk/turn/provider counts.
Caveats (same as swarm-transcript-scan)
- Status is chunk-local — expect false-OPEN inflation.
- Single-transcript blind — cross-check OPEN items against
git log, merged
PRs, and the live TodoWrite before treating any as real.
- The OPEN list is leads to verify, not a to-do list to execute.
Full documentation lives in .claude/skills/swarm-transcript-scan/SKILL.md.
