superlooperlisted
Install: claude install-skill willprout/superlooper
# superlooper — the issue-loop workflow
Work exists only as GitHub issues. Agents write them in planning conversations; **William
approves by his word** (the runner records that as an `agent-ready` label — the label is the
record, never the decision). A small deterministic **runner** — no standing LLM session
anywhere, no LLM ever in merge mechanics — takes each approved issue, builds it in one fresh
Claude session in its own worktree, and merges to the dev mainline when mechanical gates pass:
the session's tests, a real-browser drive of the changed feature, a fresh-agent review, CI
green. **Promotion (dev→prod) is William's deliberate, batched, evidence-backed decision** —
never a mechanical switch.
**This is a universal skill.** It runs on any repo through one per-repo config
(`.superlooper/config.json`); nothing repo-specific is hardcoded here. Repo-specific facts (an
eApp's `ship.sh`, its bright-line areas, its required checks) enter only through that config —
see the published `docs/ADOPTING.md` contract at `~/.claude/skills/superlooper/docs/ADOPTING.md`
(the **adopt** skill wires up a repo from scratch).
## Router — for the job in front of you, open the one reference or invoke the one sibling skill
| You are… | Go to | For |
|---|---|---|
| **writing issues** for the loop (an agent, in a planning conversation) | the **write-issue** skill (`/superlooper:write-issue`) | the rigorous issue format, the thin-issue doctrine, `type:`/`touches:`/`blocked-by`, cross-PR-promi