Report
Skill emaballarin/ccplugins/plugins/autoresearch/skills/report
Write the final summary of an autoresearch run to `./.ar/final_report.md` — trajectory, best commit and its diff, what worked, what failed, and what to try next. Use for `/ar:report`, "write up the autoresearch results", or "summarise the experiment loop". Reads `./.ar/ar.jsonl` from disk, so it works in a fresh session. Writes only the report; never iterates, commits, or reverts, and leaves the loop able to continue.From its SKILL.md
npx -y skills add emaballarin/ccplugins --skill reportAssembled from the repository path, not quoted from the project. Check it against their README if it does not work.
2 things to look at
- 3 stars3 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.
- runs commandsInstructs the agent to run 3 commands, including `cat ./.ar/ar.jsonl | tail -50` and 2 more.
SKILL.md
3.1 KB, 673 tokens by cl100k_base, as published. Nobody here has run it
/ar:report — write the run up
Turn the record into something a person can read. Do not iterate.
Steps
-
Read state (
cat ./.ar/ar.jsonl | tail -50, plusworklog.mdandresults.tsvif present). Noar.jsonlmeans nothing to report — say so and stop. -
Reconstruct the run per
${CLAUDE_PLUGIN_ROOT}/references/resume-loop.md§1, across all segments, not just the last: a segment boundary marks a harness or baseline change, so metrics either side of it are not comparable and the report must say where the boundaries fell rather than plotting through them. -
Pull the winner chain from git — it is an independent record of the same run:
git log --format='%h %s' --grep='"status":"keep"' <config.branch> git diff <config.snapshotCommit>..<bestCommit> --stat -
Write
./.ar/final_report.md:- Goal and setup — objective, metric, direction, harness, budget.
- Result — baseline → best, absolute and relative, with the confidence tier. State plainly whether the total gain clears the noise floor; a run that ended inside its own scatter did not find anything, and the report says so rather than dressing it up.
- Trajectory — the keep chain, each with commit, metric, delta and the one-line description of the change.
- What worked — the winning changes and, where the record supports it, why they worked. Distinguish an explanation from a guess.
- What failed — discarded families of hypotheses. This is the part with the most reuse value; a future run should not re-derive these.
- Caveats — crashes,
checks_failediterations, seeds that disagreed, segment boundaries. - Next — the strongest untried ideas, from
ideas.mdand the plateau history.
-
Report the path.
.gitignorecarries!.ar/final_report.md, so the report is trackable if it should be committed — offer, do not commit unasked.
Hard rules
- Write
final_report.mdand nothing else. No iterating, no commits, no branch switches, no reverts. - Every number comes from
ar.jsonlor git. Never recompute or estimate one. - Report failures and null results as prominently as wins. A loop that banked noise is a finding; concealing it wastes the next run.
Completion status
End with a terminal status token as the last line of your reply — DONE,
DONE_WITH_CONCERNS, BLOCKED, or NEEDS_CONTEXT — per
references/completion-status.md. For /ar:report, DONE once
final_report.md is written and its path reported; NEEDS_CONTEXT if ar.jsonl
is absent or empty (nothing to summarise).
What ships with it
Read from the repository
Just SKILL.md. No reference files, no scripts.