session-retrospective

To generate a retrospective from Claude Code session transcripts, the Skill parses read-only JSONL transcripts and distills them into a bounded markdown digest highlighting tool errors, retry loops, and file churn with session-id citations.

Session mining for team memory extracts evidence-backed lessons from past coding sessions, mapping role-specific insights to memory areas and routing process changes to shared workflow documentation like conventions files.

You diagnose workflow slowdowns by mining session transcripts to identify retry behaviors, tool errors, and file churn, then mapping those evidence-backed findings to candidate corrections in shared workflow docs.

You can update workflow documentation from Claude Code transcripts through an approval-gated process where the Skill proposes diffs to conventions files and advances a watermark only after you acknowledge successful writes.

Anti-memory-poisoning guardrails include read-only transcript parsing, a propose-then-wait mechanism requiring user acknowledgment before writes, and watermark advancement that occurs only after successful write operations.

The session retrospective requires no external dependencies, needing only access to Claude Code session transcripts to parse and distill into an evidence-backed markdown report for team review.