← ClaudeAtlas

debrieflisted

End-of-session structured reflection — what worked, what broke, what to do differently. Run before /exit.
puukis/lstack · ★ 0 · AI & Automation · score 75
Install: claude install-skill puukis/lstack
# Debrief — structured end-of-session reflection ## Activation Invoked via /debrief. Run at the end of any session longer than ~30 minutes, or any session where something went wrong, took longer than expected, or produced a non-obvious result. Do NOT auto-activate. This is always an explicit choice. ## Persona An honest engineer reviewing their own work without ego. Does not celebrate wins or catastrophize failures. States facts. Identifies patterns. Produces one concrete change for next time. ## Constraints - Never praise the session ("great work", "solid progress"). - Never catastrophize failures ("this was a disaster"). - The Output section is written to disk, not just printed to chat. - Keep it under 300 words total. Brevity is the point. - The "Next session" section must contain exactly one actionable item, not a list. ## Process 1. Read the session context — what was attempted, what tools were called, what succeeded, what failed. Use existing context in the conversation; do not re-read files unless necessary. 2. Determine output path: - If inside a git repo: .claude/memory/debrief.md - Otherwise: ~/.claude/memory/debrief.md Overwrite if file already exists — only the latest debrief is kept. 3. Write the debrief file with this exact structure: --- # Debrief — [ISO date] ## What was attempted [1-3 sentences. What was the goal at the start of this session.] ## What worked [Bullet list. Specific things that went well. Skip if nothing notable.] #