loop-engineering-auditlisted

# Loop Engineering Audit Skill **Loop engineering** is developing a codebase through autonomous agent loops: an agent takes a well-defined task, implements it, runs the project's own checks until they pass, and opens a PR — with no human in the inner loop. A loop only works when the project gives the agent three things: **context** it can read, **verification** it can run, and **guardrails** that make unattended mistakes cheap. This skill audits a repo for those three things across seven dimensions and produces the **ordered work plan** that closes the gaps. The audit is **read-only with exactly one write**: the finished report is printed in chat **and** saved to `docs/loop-engineering-audit.md` at the audited repo's root (`docs/` is created if missing; the file is overwritten on re-run so the project keeps one current audit). Nothing else is created, edited, committed, or pushed. ## Discovery — Do This First Classify the project before grading anything. Every check in `references/rubric.md` depends on knowing the stack and where its commands live. **1. Detect the stack(s)** from manifests at the root and in workspace globs (`package.json#workspaces`, `pnpm-workspace.yaml`, `turbo.json`, `Gemfile` in subdirs): | Signal | Stack | Where commands live | |---|---|---| | `Gemfile`, `config/application.rb` | Rails | `bin/*`, `Rakefile`, `lib/tasks/`, `.rubocop.yml` | | `package.json` (+ `tsconfig.json`) | Node / TypeScript | `package.json#scripts`, `turbo.json#tasks` | | `pyp