Initializer - Fast Worktree Setup

CRITICAL: Be fast and minimal. Your only job is:

1. Create git worktree
2. Install dependencies
3. Return the path

Do NOT analyze the codebase, find relevant files, or craft prompts. The conductor handles that.

Self-Service Mode

Workers needing context optimization should use:

Option	When to Use
prompt-enhancer agent	Get enhanced prompt via Task tool, handle wipe manually
/conductor:worker-init command	Full automated flow: analyze → craft prompt → /clear → resubmit

Flow for self-service:

1. Worker receives basic prompt from conductor
2. Worker spawns prompt-enhancer (Task tool) OR runs /conductor:worker-init
3. Gets skill-aware enhanced prompt back
4. Context reset (/clear) and resubmit enhanced prompt
5. Worker now has full context budget for implementation

See agents/prompt-enhancer.md and commands/worker-init.md for details.

Single Worktree Setup

Run this script with the issue ID and project directory:

ISSUE_ID="$1"        # e.g., TabzChrome-abc
PROJECT_DIR="$2"     # e.g., /home/matt/projects/TabzChrome

WORKTREE_DIR="${PROJECT_DIR}-worktrees/${ISSUE_ID}"
BRANCH_NAME="feature/${ISSUE_ID}"

# Create parent directory
mkdir -p "$(dirname "$WORKTREE_DIR")"

# Create worktree (try new branch, then existing, then HEAD)
if [ ! -d "$WORKTREE_DIR" ]; then
  cd "$PROJECT_DIR"
  git worktree add "$WORKTREE_DIR" -b "$BRANCH_NAME" 2>/dev/null || \
  git worktree add "$WORKTREE_DIR" "$BRANCH_NAME" 2>/dev/null || \
  git worktree add "$WORKTREE_DIR" HEAD
fi

# Install deps if package.json exists and node_modules doesn't
if [ -f "$WORKTREE_DIR/package.json" ] && [ ! -d "$WORKTREE_DIR/node_modules" ]; then
  cd "$WORKTREE_DIR"
  if [ -f "pnpm-lock.yaml" ]; then
    pnpm install --frozen-lockfile
  elif [ -f "yarn.lock" ]; then
    yarn install --frozen-lockfile
  else
    npm ci 2>/dev/null || npm install
  fi
fi

echo "WORKTREE_READY: $WORKTREE_DIR"

Output Format

Return a simple JSON object:

{
  "issue": "TabzChrome-abc",
  "worktree": "/home/matt/projects/TabzChrome-worktrees/TabzChrome-abc",
  "branch": "feature/TabzChrome-abc",
  "ready": true
}

Batch Setup

For multiple issues, run the script for each in parallel:

# Create multiple worktrees in parallel
for ISSUE in TabzChrome-abc TabzChrome-def TabzChrome-ghi; do
  (
    WORKTREE_DIR="${PROJECT_DIR}-worktrees/${ISSUE}"
    # ... same setup logic ...
  ) &
done
wait

Cleanup

Conductor handles cleanup after workers complete:

# Remove worktree (run from main repo)
git worktree remove /path/to/worktree --force
git branch -d feature/issue-id

What NOT To Do

• Do NOT read the issue details (bd show)
• Do NOT search the codebase for relevant files
• Do NOT craft worker prompts
• Do NOT analyze skills needed
• Do NOT check file sizes

The conductor does all that. You just make the worktree.