Session Retrospective

When to auto-invoke

Trigger this skill when any of the following happen during conversation:

•The user corrects a mistake ("that's wrong", "I don't like this", "this is incorrect")
•A tool call fails and requires a different approach
•Something takes multiple attempts that should have taken one
•The user shares non-obvious codebase knowledge or preferences
•The conversation is wrapping up (user says "thanks", "that's all", commits final work)

What to capture

For each lesson, record:

•What happened: Brief description of the mistake or discovery
•Why it matters: Impact on speed or correctness
•Rule: The actionable takeaway

Where to write

Actionable rules → `.claude/rules/lessons/<topic>.md`

Topic-based files under .claude/rules/lessons/. Each file stays small and focused. Current files:

•build-and-testing.md — Build commands, test execution shortcuts
•exceptions.md — Exception type usage and include patterns
•tooling.md — Claude Code tool quirks and workarounds

Create new topic files as needed (e.g., git.md, api.md, codebase.md). Format: bullet points, one rule per line, concise.

Only add a rule if it would prevent a real mistake from recurring. Do not add obvious or generic advice.

Detailed notes → memory files

Write verbose context, session-specific details, and exploratory notes to files under the auto memory directory. Use topic-based files (e.g., debugging.md, codebase-knowledge.md). Link from MEMORY.md if adding a new file.

When invoked manually

When the user runs /session-retrospective, review the full conversation and:

•Identify all corrections, failed attempts, and rediscovered knowledge
•Update the appropriate files under .claude/rules/lessons/ and memory files
•Summarize what was added or changed