hooks

<tool_restrictions>

MANDATORY Tool Restrictions

REQUIRED TOOLS — use these, do not skip:

• AskUserQuestion — Preserve the one-question-at-a-time interaction pattern for user choices. In Claude Code, use the tool. In Codex, ask one concise plain-text question at a time unless a structured question tool is actually available in the current mode. Do not narrate missing tools or fallbacks to the user.

BANNED TOOLS — calling these is a skill violation:

• EnterPlanMode — BANNED. Do NOT call this tool. This skill has its own structured process. Execute the steps below directly.
• ExitPlanMode — BANNED. You are never in plan mode. </tool_restrictions>

Step 0: Handle flags

--remove: Remove all Arc-installed hooks (Claude Code + git hooks). Read .claude/settings.json, remove Arc hooks, write back. Remove Arc-created git hooks from .husky/ or .vite-hooks/. Report what was removed. Done — skip all other steps.

--git-only: Skip Claude Code hooks (Steps 1-7). Jump directly to Step 8 (git hooks).

--claude-only: Install only Claude Code hooks (Steps 1-7). Skip Step 8 (git hooks).

No flag (default): Install both Claude Code hooks AND git hooks.

Step 1: Detect Biome

grep -q '"@biomejs/biome"' package.json 2>/dev/null

If Biome is in package.json: Continue to Step 2.

If Biome is NOT found:

AskUserQuestion:
  question: "Biome not found in package.json. Arc hooks require Biome for auto-formatting and linting. How would you like to proceed?"
  header: "Biome Required"
  options:
    - label: "Install Biome and continue"
      description: "Add @biomejs/biome as a dev dependency and set up all hooks"
    - label: "Skip formatting hooks"
      description: "Install context monitor and git guard only, skip Biome-dependent hooks"
    - label: "Cancel"
      description: "Exit without installing any hooks"

If user picks "Install Biome and continue":

# Detect package manager from lockfile
if [ -f pnpm-lock.yaml ]; then
  pnpm add -D @biomejs/biome
elif [ -f yarn.lock ]; then
  yarn add -D @biomejs/biome
else
  npm install -D @biomejs/biome
fi

Then check for biome.json / biome.jsonc. If missing:

npx @biomejs/biome init

If user picks "Skip formatting hooks": Set SKIP_BIOME=true, continue to Step 3.

If user picks "Cancel": Stop. Do not install any hooks.

Step 2: Verify Biome works

./node_modules/.bin/biome --version

If this fails, the binary isn't available. Tell the user and offer to reinstall.

Also check for biome config:

ls biome.json biome.jsonc 2>/dev/null

If no config exists, note this — biome will use defaults, which is fine.

Step 3: Build the hooks config

Build the hooks object to merge into .claude/settings.json.

Biome format hook (PostToolUse, Edit|Write):

{
  "matcher": "Edit|Write|NotebookEdit",
  "hooks": [
    {
      "type": "command",
      "command": "jq -r '.tool_input.file_path // .tool_input.filePath // empty' | { read file_path; case \"$file_path\" in *.js|*.ts|*.jsx|*.tsx|*.json|*.jsonc|*.css|*.graphql) ./node_modules/.bin/biome format --write \"$file_path\" 2>/dev/null || true ;; esac; }"
    }
  ]
}

Biome lint hook (Stop):

{
  "hooks": [
    {
      "type": "command",
      "command": "git diff --name-only --diff-filter=d HEAD 2>/dev/null | grep -E '\\.(js|ts|jsx|tsx|json|jsonc|css|graphql)$' | xargs -r ./node_modules/.bin/biome check --fix --unsafe 2>/dev/null || true"
    }
  ]
}

TypeScript check hook (Stop):

Only include this if the project has TypeScript (tsconfig.json exists).

{
  "hooks": [
    {
      "type": "command",
      "command": "npx tsc --noEmit 2>&1 | tail -20 || true"
    }
  ]
}

Git guard hook (PreToolUse, Bash):

This hook blocks destructive git operations before they execute. Always install — not dependent on Biome.

{
  "matcher": "Bash",
  "hooks": [
    {
      "type": "command",
      "command": "jq -r '.tool_input.command' | grep -qE 'git\\s+(reset\\s+--hard|push\\s+(-f|--force)|clean\\s+-f|checkout\\s+\\.)' && echo '{\"decision\":\"block\",\"reason\":\"Destructive git operation blocked by Arc hooks. Ask the user first.\"}' || true"
    }
  ]
}

Context monitor hook (PostToolUse, all tools):

Determine the absolute path to Arc's hooks/ directory by resolving the Arc install root from this skill's location. Use that resolved path as ${ARC_HOOKS_PATH}.

Read hooks/arc-context-monitor.js from the resolved Arc install root to confirm it exists.

{
  "matcher": "",
  "hooks": [
    {
      "type": "command",
      "command": "node ${ARC_HOOKS_PATH}/arc-context-monitor.js"
    }
  ]
}

Where ${ARC_HOOKS_PATH} is the resolved absolute path to the Arc plugin's hooks/ directory.

Statusline hook:

{
  "type": "command",
  "command": "node ${ARC_HOOKS_PATH}/arc-statusline.js"
}

If SKIP_BIOME is true: Only include the git guard, context monitor, and statusline hooks. The tsc hook is independent of Biome — include it if tsconfig.json exists.

Step 4: Read existing settings

cat .claude/settings.json 2>/dev/null

If file doesn't exist:

mkdir -p .claude

Start with an empty object {}.

If file exists: Parse the JSON. Preserve ALL existing fields (permissions, enabledMcpjsonServers, enableAllProjectMcpServers, enabledPlugins, etc.).

Step 5: Merge hooks into settings

This is the critical step. NEVER clobber existing hooks — merge with them.

Read the existing hooks object (may be undefined).

For PreToolUse:

• Get existing hooks.PreToolUse array (or empty array)
• Check if an Arc git guard hook already exists (command contains git.*reset.*--hard)
• Add only if it doesn't exist

For PostToolUse:

• Get existing hooks.PostToolUse array (or empty array)
• Check if an Arc biome format hook already exists (command contains biome format)
• Check if an Arc context monitor hook already exists (command contains arc-context-monitor)
• Add new entries only if they don't already exist
• Biome format hook and context monitor hook are separate entries (different matchers)

For Stop:

• Get existing hooks.Stop array (or empty array)
• Check if an Arc biome lint hook already exists (command contains biome check)
• Check if an Arc tsc hook already exists (command contains tsc --noEmit)
• Add each only if it doesn't exist
• If no tsconfig.json in the project, skip the tsc hook

For Statusline:

• Get existing hooks.Statusline array (or empty array)
• Check if Arc statusline already exists (command contains arc-statusline)
• Add only if it doesn't exist

Write the merged settings back: Use the Write tool to write the complete JSON (pretty-printed with 2-space indent).

CRITICAL: Preserve all non-hook fields exactly as they were. Do not add, remove, or modify anything outside the hooks key.

Step 6: Verify installation

Read back .claude/settings.json and confirm the hooks are present.

Count installed hooks:

• PreToolUse entries
• PostToolUse entries
• Stop entries
• Statusline entries

Step 7: Report

Arc hooks installed in .claude/settings.json

  PreToolUse  (Bash)        →  blocks destructive git ops (force push, reset --hard)
  PostToolUse (Edit|Write)  →  biome format --write on edited file
  PostToolUse (all tools)   →  context monitor (warns at 35%/25% remaining)
  Stop                      →  biome check --fix --unsafe on all changed files
  Stop                      →  tsc --noEmit (if tsconfig.json exists)
  Statusline                →  context bar showing usage

Hooks run automatically — zero token cost, zero agent effort.

To remove: /arc:hooks --remove

If SKIP_BIOME was true:

Arc hooks installed in .claude/settings.json

  PreToolUse  (Bash)        →  blocks destructive git ops (force push, reset --hard)
  PostToolUse (all tools)   →  context monitor (warns at 35%/25% remaining)
  Statusline                →  context bar showing usage

Biome hooks skipped (not installed). Run /arc:hooks again after adding Biome.

To remove: /arc:hooks --remove

Step 8: Install git hooks (Husky or Vite+)

This step installs pre-commit and pre-push git hooks to enforce typecheck + lint at the git level. This catches errors that Claude Code hooks can't — like commits made outside Claude, or when the AI session ends before running Stop hooks.

Step 8a: Detect hook system

Check which hook system the project uses:

1. Vite+ project: .vite-hooks/ directory exists, OR package.json has "prepare": "vp config" → use Vite+ hooks
2. Husky project: .husky/ directory exists, OR package.json has "prepare": "husky" → use Husky hooks
3. Neither: Install Husky (see Step 8b)

Step 8b: Ensure hook infrastructure exists

For Husky projects (or new installs):

Check if husky is installed:

grep -q '"husky"' package.json 2>/dev/null

If not installed:

# Detect package manager
if [ -f pnpm-lock.yaml ]; then
  pnpm add -D husky lint-staged
elif [ -f bun.lockb ] || [ -f bun.lock ]; then
  bun add -D husky lint-staged
elif [ -f yarn.lock ]; then
  yarn add -D husky lint-staged
else
  npm install -D husky lint-staged
fi

Ensure prepare script exists in package.json:

"scripts": { "prepare": "husky" }

Ensure .husky/ directory exists:

mkdir -p .husky

For Vite+ projects: Hooks go in .vite-hooks/. Ensure directory exists:

mkdir -p .vite-hooks

Step 8c: Configure lint-staged

Check if lint-staged config exists in package.json. If not, add it:

"lint-staged": {
  "*.{js,ts,jsx,tsx,json,jsonc,css}": "biome format --write --no-errors-on-unmatched"
}

For Vite+ projects: lint-staged is handled by vp staged configured through vite.config.ts. Do not add lint-staged config to package.json. If vp staged is not configured, check vite.config.ts for staged config. If missing, note this to the user.

Step 8d: Create hook files

Determine the hooks directory ($HOOKS_DIR):

• Husky: .husky/
• Vite+: .vite-hooks/

pre-commit ($HOOKS_DIR/pre-commit):

For Husky projects:

pnpm lint-staged
pnpm typecheck && pnpm lint

For Vite+ projects:

vp staged

For bun-based projects (detected by bun.lockb or bun.lock):

bun run typecheck && bunx lint-staged

pre-push ($HOOKS_DIR/pre-push):

For Husky projects:

pnpm typecheck || exit 1
pnpm lint || exit 1

For Vite+ projects (adapt to the project's existing scripts):

pnpm typecheck || exit 1
pnpm lint || exit 1

Make both files executable:

chmod +x $HOOKS_DIR/pre-commit $HOOKS_DIR/pre-push

Important: If hook files already exist, read them first. Only overwrite if they are missing typecheck or lint steps. Never remove project-specific steps that are already present (like env validation, SDK guards, test runs, etc.).

Step 8e: Fix turbo.json cache (Turborepo projects only)

If turbo.json exists, ensure typecheck and lint tasks have cache: false:

# Read turbo.json and check cache settings

For each task (typecheck, lint, check-types):

• If the task exists and cache is not false, set cache: false
• If the task does not exist, skip it (don't add tasks that aren't defined)

This is non-negotiable. Turborepo caching on typecheck/lint is the #1 cause of type errors slipping through git hooks. Turbo replays a cached success exit code without actually running tsc or biome, making hooks pass when they should fail.

Do NOT disable cache on build or test tasks — those have legitimate outputs and benefit from caching.

Step 8f: Report git hooks

Git hooks installed in $HOOKS_DIR/

  pre-commit  →  lint-staged (biome format on staged files) + typecheck + lint
  pre-push    →  full typecheck + lint (safety net before push)
  turbo.json  →  cache: false for typecheck/lint (if applicable)

Errors will now be caught before they leave your machine.

## Claude Code hooks - The biome format hook uses `jq` to extract the file path from the tool input. This is standard on macOS (via Homebrew) and most Linux distros. If jq is missing, the hook silently fails (|| true). - The Stop hook uses `xargs -r` which is a no-op if there are no files. On macOS, `xargs` without `-r` still works (just runs biome with no args, which exits cleanly). - The context monitor hooks reference files inside the Arc plugin. If the user uninstalls Arc, these hooks will silently fail (node script not found → exit 1, but hooks don't block). - `--unsafe` in the Stop hook enables Biome's unsafe fixes (like removing unused imports). This is intentional — at conversation end, we want maximum cleanup. - The PostToolUse format hook handles both `file_path` (Edit tool) and `filePath` (NotebookEdit) via jq fallback. - The tsc hook uses `tail -20` to avoid flooding the stop output — just enough to see if there are errors and what they are. - The git guard uses PreToolUse with `decision: block` to stop destructive commands before execution. This is especially important for users running with `--dangerously-skip-permissions` or liberal auto-approve settings. - The git guard blocks: `git reset --hard`, `git push --force` / `git push -f`, `git clean -f`, `git checkout .`. These are the operations that destroy uncommitted work with no undo.

Git hooks

• Husky v9 uses a flat .husky/ directory — hook files are directly in .husky/pre-commit, .husky/pre-push, etc. No _ subdirectory or husky.sh sourcing needed.
• Vite+ uses .vite-hooks/ with a _/ subdirectory containing the hook runner. git config core.hooksPath is set to .vite-hooks/_ by vp config. User hooks go directly in .vite-hooks/pre-commit, .vite-hooks/pre-push, etc.
• The pre-commit hook runs pnpm lint-staged first (fast — only formats staged files), then pnpm typecheck && pnpm lint (catches errors before commit).
• The pre-push hook runs the same typecheck + lint as a safety net. It uses || exit 1 instead of && so each step reports its own failure.
• Turborepo cache: false is the most important fix. Without it, turbo typecheck in a git hook may return a cached success from a previous run, silently passing despite new type errors. This is the #1 cause of type errors getting through hooks.
• cache: false should ONLY be set on typecheck, lint, and check-types tasks. build and test tasks should keep their cache — those have real outputs and benefit from caching.
• Never remove project-specific pre-push steps (env validation, SDK guards, case-sensitivity checks, unit tests). Only add missing typecheck/lint steps.