wf-setup

WF: Team Setup

Gate Check

cat .claude/.workflow-state/plan-complete.json

If the file does not exist: STOP — "Plan not loaded. Run Step 2 (wf-plan) first."

Load: TICKET, TOTAL_WAVES, waveMap, task file list. Also reload preflight-complete.json for PLUGIN_ROOT, mode, currentBranch, featurePrefix, workPrefix, worktreeDir, useWorktrees.

---

Your Task

Create the team, prepare task files with runtime values, create worktrees, and inject a CLAUDE.md into each worktree.

---

Checklist

1. TeamCreate

TeamCreate: team_name = "<TICKET>"

After creation, read team config to get your own name:

cat ~/.claude/teams/<TICKET>/config.json

Extract your TEAM_LEADER_NAME from the members array. Store it.

2. Clean Stale State

git worktree list --porcelain \
  | grep "worktree.*<worktreeDir>/<TICKET>" \
  | awk '{print $2}' \
  | xargs -r -I{} git worktree remove --force {}

git branch --list "<workPrefix>/<TICKET>/*" \
  | xargs -r git branch -D

3. Update Task Files with Runtime Values

For each task-*.md file, update YAML frontmatter:

• teamLeaderName → TEAM_LEADER_NAME
• teamName → TICKET
• workbranch → <workPrefix>/<TICKET>/<taskSlug>
• worktreePath → <worktreeDir>/<TICKET>/<taskSlug>
• status → "pending"

4. TaskCreate with Dependencies

For each task, call TaskCreate with description + acceptance criteria from the task file. Then call TaskUpdate with addBlockedBy for each entry in the task's blockedBy field.

5. Create Worktrees (Wave 1 only)

For each task in Wave 1:

git worktree add <worktreeDir>/<TICKET>/<taskSlug> \
  -b <workPrefix>/<TICKET>/<taskSlug>

If useWorktrees is false, create branches instead and continue with sequential execution.

6. Inject CLAUDE.md into Each Worktree

For each Wave 1 task worktree, write <worktreeDir>/<TICKET>/<taskSlug>/CLAUDE.md:

# Task #{taskNumber}: {taskName}

You are **{agentRole}** on team "{teamName}".
Workbranch: `{workbranch}`.

## Agent Protocol

{paste full contents of agents/{agentRole}.md — minus YAML frontmatter}

## Task Requirements

{paste task file body: acceptance criteria, file scope, rules,
implementation notes}

## Workflow Phases

Read `{PLUGIN_ROOT}/prompts/implementing-features/AGENT-WORKFLOW-PHASES.md`
and follow Phases 1–4.

## Communication

- Report ONLY to "{TEAM_LEADER_NAME}" via SendMessage.
- Do NOT message other agents. Do NOT spawn agents.
- On completion: SendMessage(to: "{TEAM_LEADER_NAME}",
  message: "Task #{taskNumber} complete. Files: <list>. Self-review passed.")
- On blocker: message leader immediately.
- Wait for shutdown_request when done.