Trace

Run at the end of a session the user wants to close, before /clear. It leaves a dated save that next offers to resume. Trace never resumes work itself. Resolve scripts/trace.cjs from this skill's directory.

Save (no argument, or save)

1. Gather, without asking: this conversation; git log since the last save (node scripts/trace.cjs saves --all gives its date); git status --short; planned work: briefs (docs/brief/*/brief.md) and proposals (docs/proposals/*/PROPOSAL.md) not shipped or superseded, review reports (docs/review/*/report.md) with unchecked points, and open issues from gh issue list --label arsenal --state open when gh answers.
2. Show at most five lines: done, not finished, planned.
3. Ask only what the context cannot answer, one question per message, in plain text and the conversation language, then wait:
   • what to continue next session, with your suggestion;
   • anything to remember that you may have missed: a decision, an approach that failed. Skip a question the conversation already answers. Usually none is needed.
4. Write it: node scripts/trace.cjs save --done "…" --left "…" --next "a ; b" --remember "…" --planned "…". One line per field, in the conversation language, specific: paths, branch, task ids, issue numbers, commands. Omit an empty field.
5. End with « Tu peux /clear. » in the conversation language, then the resume command as the last line: next in the harness syntax (/next:next in Claude Code, /skill:next in OMP and Pi, $next in Codex).

The save is private: the script adds the ledger to .git/info/exclude and it never enters a commit.

Other modes

• done "<what>" [--files a,b] [--id id] [--status shipped]: one intent entry. shipped only with evidence.
• --read n: print the last entries, then stop. Do not reopen the work they describe.

The Stop hook logs files moved and commits made each turn, with no model call. It records activity, not delivery: never upgrade wip to shipped without evidence. Keep this cross-session ledger distinct from a ship run's trace.md; do not copy events between them.

Execution policy

Work solo. Ask before any subagent or reviewer, even in auto mode. No hidden advisor, nested delegation, model retuning, repeated successful checks, or progress spam. Stop when the save is written. User stops and scope changes override pending steps.