loopwrightlisted
Install: claude install-skill R4wraith/Loopwright-SEC
# Loopwright
Turn any project idea into a **drop-in `.claude/` folder** that makes Claude Code build it autonomously, task after task, as a lead-engineer/PM that dispatches specialist subagents through a persistent loop — with an event-sourced run ledger, operator shifts, safety hooks, git discipline, and an honest build journal baked in.
The value of this skill is the **tailoring**, not the file copy. Most of the harness is reusable as-is; the win is doing the architecture thinking for *this* idea and generating the right design doc + subagent roster. Don't skip the thinking steps.
## What it produces
A single `.claude/` folder the user drops into their project root and runs `/start`:
- `CLAUDE.md` — the constitution (how the agent works, ≤60 lines; rationale lives in `references/blueprint.md`) — auto-loaded by Claude Code from `.claude/`. A repo that already has a root `CLAUDE.md` gets a short `## Loopwright` section appended there instead (≤12 lines full, ≤8 lite) (one constitution per repo; its conventions win).
- `WORKFLOW.md` — the **verbatim** run → shift → iteration → slice mechanism (`Harness-Version: 3.0`). `hooks/workflow-state.mjs` parses its `[workflow-state:*]` blocks and injects the one matching the current status **every turn**, with an exact-resume pointer.
- `DESIGN.md` — what's being built (architecture, keystone, build order).
- `GOAL.md` / `STATE.md` / `PROGRESS.md` / `DECISIONS.md` / `FINDINGS.md` / `LEARNINGS.md` — pre-seeded mission, state, honest l