| name | dev-loop-setup |
| description | Scaffold a Rails project to use /dev-loop. Verifies @kurenn/roundhouse and @openai-codex/codex are installed, adds a '## Dev-loop config' block to CLAUDE.md (gate, critical paths, extra rating axes, learnings path, PR conventions), and creates the learnings file. Run once per repo before the first /dev-loop. |
/dev-loop-setup — prepare a repo for the loop
One-time setup. Make the smallest set of changes that lets /dev-loop run cleanly in
this repository, then tell the user what (if anything) they still need to do.
Step 1 — Check prerequisites
Report whether each is installed; don't fail if missing (the loop degrades), but tell
the user what they lose:
ls -d ~/.claude/plugins/cache/kurenn/roundhouse/*/ 2>/dev/null
ls -d ~/.claude/plugins/cache/openai-codex/codex/*/ 2>/dev/null
- Missing roundhouse → Phase 2 falls back to directly-spawned Sonnet agents. Recommend
/plugin install roundhouse@kurenn.
- Missing codex → Phase 3 falls back to an Opus adversarial reviewer agent. Recommend
/plugin install codex@openai-codex.
Step 2 — Scaffold the ## Dev-loop config block in CLAUDE.md
If the repo has no CLAUDE.md, create one. If it has one but no ## Dev-loop config
section, append the block below. If the section already exists, leave it and show the
user the current values. Fill the placeholders by inspecting the repo — don't leave
<...> in the file:
- Critical paths: infer from the domain (e.g. a fintech app → money / auth / KYC /
compliance; a healthcare app → PHI / auth). If none are obvious, write
none.
- Extra rating axes: add
design-system fidelity if the repo has a component
library / design system; otherwise none.
- PR conventions: capture any repo-specific quirks you can detect (private repo →
screenshot embedding workaround; broken
gh pr edit; assets-branch pattern).
## Dev-loop config
Knobs read by the `/dev-loop` skill (@kurenn/dev-loop). Defaults shown; edit to taste.
- **Gate:** overall ≥ 8.5 AND every axis ≥ 7
- **Base rating axes:** correctness, simplicity, test coverage, naming, performance risk, security risk
- **Extra rating axes:** <e.g. design-system fidelity — or `none`>
- **Critical paths:** <e.g. money / auth / KYC — these axes also require ≥ 8.5 — or `none`>
- **Learnings file:** docs/dev-loop-learnings.md
- **Fix-round cap:** 2
- **PR conventions:** <repo-specific screenshot/embed/gh quirks — or `standard gh pr create`>
Step 3 — Create the learnings file
If the configured learnings file (default docs/dev-loop-learnings.md) doesn't exist,
create it with this header so Phase 6 has somewhere to append:
# Dev-loop learnings
Durable, reusable insight captured by `/dev-loop` (Phase 6). Append here — newest first.
Map, not diary. Nothing user-specific or secret.
## Entry format
### YYYY-MM-DD — <short title>
**Context:** <feature/bug + PR # if any>
**Learning:** <the durable, reusable insight>
**How to apply:** <what to do next time>
Step 4 — Report
Tell the user: what was created/edited, which prerequisites are missing (with the exact
/plugin install commands), and that they can now run /dev-loop <task>.