shipshitdev__skills/commands/loop.md

Loop — Execute one prepared issue

Run the executing-plans skill for one explicitly authorized dispatch:claude issue. This is the local Claude entry point; the selected harness owns model, effort, account, checkout and approval configuration.

Usage

/loop            Execute one eligible prepared issue
/loop --status   Show current ownership and delivery state; read-only
/loop --list     List eligible queue candidates and blockers; read-only

Contract

Resolve executing-plans through the active skill catalog and read its installed references/delivery-gate.md. Resolve prd-quality-gate and its installed references/execution-readiness.md the same way. If the repository has the provisioned .github/agent-dispatch.md, read that shared dispatch contract too. Missing required resources block execution; never reconstruct a weaker contract from this command or assume the consumer contains this source repository.

• For --status or --list, inspect only. Do not claim, comment, edit or dispatch.
• Otherwise, process exactly one candidate with the selected gate and verified Backlog state, satisfied dependencies and a current prepared plan. Run the blocking semantic execution-readiness check before editing. A label or an old READY marker alone is insufficient.
• Serialize ownership across provider lanes. Confirm the run ended before explicit claim recovery; elapsed time never authorizes takeover. If ownership cannot be verified, stop intake. An already claimed push-workflow run owns its own claim.
• Implement the selected issue revision without making product or engineering decisions. Return missing or contradictory decisions to the planner.
• Publish the PR with Refs #<issue> and the current plan link/revision. Record acceptance evidence and the actual implementation provider, then hand off as review_pending. Self-QA and reviewer assignment do not satisfy independent review.
• Done requires independent review by a different model provider/lab from every implementation contributor, green required CI at the reviewed final head, verified merge, and required deployment/migration/smoke evidence. Keep the epic open until every required child outcome and the integrated feature are complete.
• Follow the engine's ownership and delivery receipt rules; release only this run's claim after a resumable handoff. This command grants no additional merge, deployment or provider-switching authority.

Preparation belongs to /prd prepare or the authorized dispatch:plan OpenAI planner workflow. Preparation does not apply an execution gate. Use the harness's configured executor to run this contract; keep lifecycle procedures in the shared engines instead of copying them into a CLI prompt.