Author a Crit story

Author the story in-session, then continue the interactive Crit review loop. Do not run bare crit story unless the user asks for the agent_cmd path.

1. Read the guide and diff

crit story --guide
crit story --prep /tmp/crit-story-prep.txt

The guide prints the user's resolved authoring instructions followed by --- and the JSON schema. Follow it rather than assuming this skill contains the current guide. Read the prep file for the full diff and its (file_path, old_start) hunk identifiers.

2. Author and ingest

Use the write tool to write /tmp/crit-story.json. Cluster hunks by theme, not by file, following the guide. Emit only prologue, chapters, and support; Crit fills in metadata and coverage.

crit story --story-file /tmp/crit-story.json

Inspect the JSON coverage report on every attempt. Exit 1 means rejection: fix duplicate or missing hunk assignments and re-ingest. A drift error means the diff changed; fetch fresh prep and re-author. Exit 0 with auto_repaired: true means omissions were back-filled into support.

3. Reconnect and await Finish Review

Ingest opens the browser and exits; it does not wait for the human. Run bare crit through omp's bash tool as an asynchronous job:

{"command":"crit","async":true,"timeout":0}

Save the returned job ID. timeout: 0 disables the default five-minute command deadline; async: true alone does not. If asynchronous jobs are unavailable, use foreground bash with timeout: 0. Do not use a named service for this finite review-round command.

Relay the review URL verbatim; if omitted from startup output, read proc://<job-id> once. Use wait when no independent work remains. Completion also auto-delivers. A running snapshot or unrelated message is not completion: keep waiting until this job exits. Do not poll process output or the review file, hold an eval cell open, read feedback early, or ask the user to type a reply to signal Finish Review.

4. Address feedback and repeat

Read the command output and follow the finish prompt; check for approved: true or approved: false. omp merges stdout and stderr. Retrieve omitted output through proc://<job-id> or its artifact:// link. A command failure is not approval; report the error rather than proceeding. If approved, stop. Otherwise read every unresolved comment and its replies, including review-level comments. Use quote and anchor rather than stale line numbers when edits have moved the content.

Address comments against the actual source files. Do not regenerate or edit the saved story JSON unless the user explicitly asks.

crit comment --reply-to <id> --author 'omp' '<what changed>'
crit comment --json --file /tmp/crit-story-replies.json --author 'omp'

Write bulk reply JSON with the write tool; each entry has reply_to and body. Do not use --resolve unless explicitly requested. For mid-round re-entry, use crit comments --json.

Run the next-round command printed on stdout with the same asynchronous bash pattern and timeout: 0. Tell the user only that you replied in Crit and the changes are ready for review. Do not repeat your replies or list the comments in chat. Await command completion, and repeat this step until approved. Do not call crit push or crit share unless the user asks.