crit-story
Skill4 starsChanged 16 days agoApache-2.0View on GitHub
Author a crit story and continue the interactive review loop only when the user explicitly invokes $crit-story or directly asks you to generate a crit story. Do not infer this skill from generic review, PR, or diff-review requests.
In this repository10 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- bump-versionSkill
- check-changesSkill
- check-versionSkill
- crit-cliSkill
- critSkill
- bump-versionSkill
- check-changesSkill
- check-versionSkill
The file
We split this skill into the rules it gives an agent
8 rules. About 0 are common, 2 are distinctive, and 6 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
6
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 77
If ingest printed a review URL, relay it verbatim:
- Very specific: line 81
Do NOT proceed until
critcompletes: Do NOT ask the user to type anything. - Very specific: line 87
When crit completes, read stdout and follow its instructions.
- Very specific: line 89
If the finish prompt says the review is approved / "approved": true, tell the user no changes were requested and stop.
- Very specific: line 91
When a comment has quote, anchor, or drifted:
- Distinctive: line 105
If it contains a suggestion block, apply that specific change
- Very specific: line 116
Use --json for a single bulk call instead of one invocation per comment:
- Very specific: line 141
Do NOT proceed until the background
crittask completes: When it does, return to Step 6.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Author a crit story with crit story
This skill authors a story-mode overview, then enters the same interactive
review cycle as /crit (wait → address comments → next round). Invoke it
only when the user explicitly asks for story generation; do not infer it from
generic review, PR, or diff-review requests.
Primary path: author in-session with --guide / --prep / --story-file
(do not run bare crit story, which spends agent_cmd tokens), then run
bare crit to listen for Finish Review.
Step 1: Fetch the guide
crit story --guideThis prints the resolved authoring guide (the user's customized version if
they have one — always invoke this at runtime rather than reusing this
skill's own prose) followed by --- and the JSON schema for the fields you
must emit. Read and follow that guide's principles and JSON shape exactly —
it is the source of truth, not this file.
Step 2: Write the prep file
crit story --prep /tmp/crit-story-prep.txtThis writes the full, untrimmed diff (commit messages + every hunk with its
(file_path, old_start) id) to the given path and prints the path. Read
that file — the diff is never inlined into the guide prompt.
Step 3: Author the story JSON
Following the guide from Step 1, cluster hunks by theme (not by file) and
write a JSON object with only prologue, chapters, and support to a
temp file, e.g. /tmp/crit-story.json. Do not include version,
generated_at, agent, base_sha, head_sha, scope_fingerprint, or
coverage — crit fills those in.
Step 4: Ingest
crit story --story-file /tmp/crit-story.jsonExit 0 means the story was saved (crit opens the browser to show it). Then continue to Step 5 — do not stop after ingest. Exit 1 means it was rejected — the coverage report (missing/duplicated hunks) is printed to stdout as JSON on every attempt, success or failure. If rejected:
duplicatednon-empty: a hunk is claimed by two chapters (or a chapter and support). Decide where it belongs and re-ingest.missingnon-empty withauto_repaired: trueon exit 0: crit already back-filled the omissions intosupport[]. Optionally re-author to place them deliberately.- A drift error ("diff changed since prep"): re-run
crit story --prepand re-author from the fresh prep file.
Step 5: Reconnect and wait until Finish Review
CRITICAL — you MUST run this step after a successful ingest. Do NOT skip it. Do NOT proceed without it.
Story ingest opens the review UI and exits — it does not wait for the human. Run crit and wait until it exits:
critcrit reconnects to the story session's daemon (already started by ingest) and blocks until the user clicks "Finish Review".
If ingest printed a review URL, relay it verbatim:
"Crit is open at http://localhost:
<port>. Leave inline comments on the diff, then click Finish Review."
Do NOT proceed until crit completes. Do NOT ask the user to type anything. Do NOT read the review file early. Wait for the command to finish — that is how you know the human is done reviewing.
While crit is still running, do not send user-facing progress updates. Treat the wait loop as internal state.
Step 6: Read the review output
When crit completes, read stdout and follow its instructions. Check stderr for approved: true or approved: false.
If the finish prompt says the review is approved / "approved": true, tell the user no changes were requested and stop.
When a comment has quote, anchor, or drifted:
quote: the specific text the reviewer selected — focus your changes on the quoted text rather than the entire line rangeanchor: use it to locate the current position of the content; line numbers may be stale after editsdrifted: true: original content was removed or heavily rewritten — line numbers are approximate at best
Fallback (mid-round re-entry or headless workflows): crit comments / crit comments --json.
Step 7: Address each review comment
Story mode is an editorial view over the underlying diff. Address comments against the actual source files. Do not edit, regenerate, or patch the saved story JSON unless the user explicitly asks.
For each unresolved comment:
- Understand what the comment asks for
- If it contains a suggestion block, apply that specific change
- Revise the referenced source file
- Reply with what you did:
crit comment --reply-to <id> --author 'Codex' '<what you did>'(reply bodies support markdown) - Do not pass
--resolve. Resolving is the reviewer's call. Only add--resolveif the user explicitly asks.
<important if="you are revising in plan mode">
Re-emit the revised plan inside `<proposed_plan>...</proposed_plan>` so Crit can review the new version.
</important>
### When replying to multiple commentsUse --json for a single bulk call instead of one invocation per comment:
echo '[
{"reply_to": "c_a1b2c3", "body": "Fixed"},
{"reply_to": "c_d4e5f6", "body": "Refactored as suggested"}
]' | crit comment --json --author 'Codex'For long or multi-line bodies, write the JSON to a file and pass --file. Use --file, not a < redirect.
crit comment --json --file /tmp/replies.json --author 'Codex'Step 8: Signal completion and start next round
CRITICAL — you MUST run this step. Do NOT skip it. Do NOT proceed without it.
The finish prompt on stdout includes the command to run again — use it to start a new round.
On subsequent calls, crit automatically signals round-complete first, then blocks until the next "Finish Review" click.
Tell the user only: "Replied in Crit. Review the changes in your browser and click Finish Review when ready." Do not repeat your replies or list the comments in chat. The reviewer reads them in Crit.
Do NOT proceed until the background crit task completes. When it does, return to Step 6. If the user finishes with zero comments, the review is approved — stop the loop and proceed.
What this skill does NOT do
- It does not produce agent-authored review comments during story authoring — humans leave comments in the browser after ingest.
- It does not edit or regenerate the saved story JSON during the review loop unless the user explicitly asks.
- It does not run bare
crit story(LLM viaagent_cmd) unless the user explicitly asks for that path. - It does not call
crit pushorcrit shareunless the user asks to share.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

