coding-quality-loop

One-command installer for the Coding Quality Loop skill — an engineering operating system that makes AI coding agents ship small, verified changes you can trust.

Install

npx coding-quality-loop init

That's it. The installer auto-detects your host, copies the skill files, wires the hooks, and prints next steps. No prerequisites beyond Node 18+ and Python 3. The routed loop runs Claude Code (implementer) + Codex (independent reviewer); Droid is a supported install target outside that two-vendor kernel.

Commands

npx coding-quality-loop init              # interactive install
npx coding-quality-loop init --yes        # accept all defaults
npx coding-quality-loop init --host codex # skip host detection
npx coding-quality-loop init --dry-run    # preview without writing
npx coding-quality-loop add git           # add a single host later
npx coding-quality-loop check             # verify a prior install against its manifest
npx coding-quality-loop remove            # uninstall: remove installed files, reverse hook wiring
npx cql --help                            # short alias also works

Every install writes a manifest at .quality-loop/install-manifest.json recording each file written and each hook group merged. cql check verifies the install against that manifest, and cql remove uninstalls from it — restoring any .bak backups of your pre-install files and leaving your own files alone.

What it installs

Depending on the host you pick:

• Claude Code — copies the skill (SKILL.md, references/, assets/) to .claude/skills/coding-quality-loop/, vendors the runtime scripts once at scripts/, wires .claude/settings.json hooks (PreToolUse, Stop, SessionStart), and installs review/security-review subagents.
• Codex — copies AGENTS.md (created from the bundled template; if you already have one, a clearly-marked managed section is appended instead) and wires .codex/hooks.json.
• Droid — copies role droids into .factory/droids/.
• Git backstop (recommended) — installs a pre-commit hook that runs diff-audit --staged to block secrets, weakened tests, and untracked-file leaks before commit.

Cursor and Pi are not runtime install targets: advisory rules recipes for them live in the repository's examples/ directory.

All installs are idempotent, non-destructive (your pre-existing files are backed up as .bak), and can be previewed with --dry-run.

After install

# 0. Commit the install so the diff gates start from a clean base
git add -A && git commit -m "chore: install coding-quality-loop"

# 1. Create your config from the installed example
cp assets/quality-loop.config.example.json quality-loop.config.json

# 2. Configure per-role model routing once
python3 scripts/quality_loop.py setup-models --host claude-code

# 3. Try it out
claude "Use coding-quality-loop to fix a small bug"

Docs

Full documentation, philosophy, and eval results are in the main repository.

License

MIT