setup

name: setup description: Use when the user asks to set up AutoReview in the repo ("setup autoreview", "init autoreview", "initialize autoreview", "bootstrap autoreview", "add autoreview here", "install autoreview") or when .autoreview/ is missing and the user wants code review. Also triggers when the agent notices no config but rules are being discussed. Skip when .autoreview/ already exists and the user is not asking for a reset (use autoreview:create-rule instead to add rules).

AutoReview Setup

Run when the repo has no .autoreview/ directory and the user wants to enable rule-based LLM review.

Cross-platform. Snippets below use bash-style env-var syntax (${CLAUDE_PLUGIN_ROOT}). Claude Code's Bash tool runs Git Bash on Windows so it works there too. On native PowerShell substitute $env:CLAUDE_PLUGIN_ROOT; on cmd use %CLAUDE_PLUGIN_ROOT%. The plugin requires Node ≥22 — that's the only assumed binary. Run node -e "console.log(process.platform)" to detect OS ('win32' / 'linux' / 'darwin') when it matters.

Steps

1. Probe for Ollama. Use Node so it works on every platform without curl:

   node -e "fetch('http://localhost:11434/api/tags',{signal:AbortSignal.timeout(2000)}).then(r=>process.exit(r.ok?0:1)).catch(()=>process.exit(1))"
   

   Exit 0 → reachable, recommend ollama as the default provider. Exit 1 → unreachable; list provider options (ollama, claude-code, codex, gemini-cli, anthropic, openai, google, openai-compat) and ask which one to use. Do NOT ask the user to paste an API key in conversation — the secrets file is a placeholder only.

2. Model choice (ollama only). If the user picks ollama, list installed models:

   node -e "fetch('http://localhost:11434/api/tags').then(r=>r.json()).then(j=>console.log(j.models.map(m=>m.name).join('\n')))"
   

   Ask which one to use. Do NOT assume the default qwen2.5-coder:7b is pulled. After init, patch .autoreview/config.yaml's tiers.default.model to the chosen model (or tell user to run ollama pull <model> if they want a different one).

3. Pre-commit hook? Ask "install git pre-commit hook?" before init. If yes, add --install-precommit. Never install without explicit user okay.

4. Run init.

   node ${CLAUDE_PLUGIN_ROOT}/scripts/bin/init.mjs --provider <chosen-provider> --install-precommit
   

   Supported flags: --upgrade (refresh existing config), --provider <name>, --install-precommit. Omit --install-precommit if the user declined in step 3.

5. Wrap-up. Mention that the git pre-commit hook was installed (if requested) and the first example rule lives at .autoreview/rules/example.md. Briefly explain two key concepts the generated config.yaml comments cover in detail:

   • Tier model — rules declare which tier they need (tier: trivial, tier: standard, etc.). Each tier maps to a provider+model in config.yaml. Changing providers is a config edit, not a rule rewrite. Commented tier examples are already in the generated config.
   • Overlay model — if the repo pulls remote rule packs, individual rules can be tuned per-repo via remote_rules[].overrides without forking upstream. Useful for downgrading a noisy corp rule to severity: warning, or narrowing its triggers.

   Point the user at the autoreview:create-rule skill to author their first rule. For remote rule overlays, use autoreview:override-rule.

Never read .autoreview/config.secrets.yaml — it may contain keys.