wraplisted

# `/wrap` — Session-end retrospective Run this at the end of a session. It captures what was learned, evaluates how the user's tooling stack performed, and stages the next session. ## What you produce 1. **Dated retro file** at `~/.claude/sessions/YYYY-MM-DD-<cwd-slug>-<sessionid8>.md` — the canonical record. 2. **Tooling journal entry** appended to `~/.claude/tooling-journal.md` via `journal-append.ts`. 3. **`NEXT_SESSION.md`** at the project root — written, left alone, or removed per the lifecycle rules below. 4. **(Conditional) CLAUDE.md edits** — user-level or project-level, only with explicit user confirmation. ## Procedure ### 1. Run `scan.ts` The skill's lib scripts live at the plugin root, two levels up from this SKILL.md. Use the base directory Claude told you about for this skill: ```bash bun run "<skill-base-dir>/../../lib/scan.ts" --cwd "$(pwd)" ``` Parse the JSON. If `ok: false`, note `degraded: true, reason: "..."` and proceed with self-reported stats from your own session memory. **Buffering note:** Claude Code may buffer transcript writes, so the last few events (including the `/wrap` call itself) may not appear in the count. If `session_end` from `scan.ts` is more than ~60s behind the current wall-clock time, mention "stats trail by ~Ns" in the journal entry under the relevant tool — otherwise the count silently undercounts. **Compaction note:** If `compaction_count > 0`, the current jsonl is a post-compaction segment and `turn_count` counts ONLY tu