session-retrolisted

This skill practices **compound engineering**: every successful session should leave the toolset slightly better. It is the tooling cousin of "end every session by updating CLAUDE.md so the same mistake never recurs." Skills live in this dotfiles repo at `claude/skills/<name>/SKILL.md` and are **symlinked** into `~/.claude/skills/`, so they sync across machines — any change must be **committed** here to propagate. CLAUDE.md / memory live alongside; architectural learnings may instead warrant an ADR in `docs/adr/`. ## Scope — what this skill is and is NOT - **IS:** improving the *toolset* — skills, their trigger descriptions, CLAUDE.md, ADRs. - **IS NOT:** logging what happened. For that, defer to the existing skills: - `changelog` — what changed this session - `log-error` — a persistent error pattern + classification - `handoff` — context for the next session If the user wants a record, point them there and stop. ## When NOT to run - Mid-task thank-yous, or a "thanks" before the work is actually done — wait for a real stopping point. - Trivial / social pleasantries with no session learning behind them. - A session where nothing notable was learned — say so in one line and stop (see step 5). - A session you already retro'd — never retro the same session twice. - When the user explicitly just wants a log entry — route to `changelog` / `log-error` / `handoff`. ## Workflow 1. **Reflect silently.** Scan THIS session for toolset-level signals, e.g.: - A skill that