wf-spawn

WF: Spawn Wave

Gate Check

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

If the file does not exist: STOP — "Team setup not complete. Run Step 3 (wf-setup) first."

Load: TICKET, TEAM_LEADER_NAME, currentWave, waveMap, PLUGIN_ROOT, workPrefix, worktreeDir. Also check for a wave-specific gate:

# If currentWave > 1, previous wave must be complete
cat .claude/.workflow-state/wave-$((currentWave - 1))-complete.json 2>/dev/null

If previous wave stamp missing and currentWave > 1: STOP — "Wave <N-1> not complete. Run wf-qa-gate for that wave first."

---

Your Task

Spawn agent pairs for the current wave. Both the coder and QA agent are spawned in the same message (parallel tool calls).

---

Checklist

1. Identify This Wave's Tasks

From waveMap, get the list of task slugs for currentWave. For each slug, read the corresponding task file to get: taskNumber, taskName, taskSlug, agentRole, agentDefinition, workbranch, worktreePath.

2. Read Spawn Templates

cat {PLUGIN_ROOT}/prompts/implementing-features/THIN-SPAWN-TEMPLATE.md

Use the Coding Agent and QA Agent templates from this file.

3. Spawn Pairs (one message per task, both agents in parallel tool calls)

For each task, in a single message, call the Agent tool twice:

Coder:

Agent tool:
  description: "<taskSlug> coder"
  subagent_type: general-purpose
  team_name: "<TICKET>"
  name: "coder-task-<N>"
  mode: bypassPermissions
  model: sonnet
  run_in_background: true
  isolation: worktree  (only if useWorktrees is false)
  prompt: <substitute THIN-SPAWN-TEMPLATE.md coding template>

QA:

Agent tool:
  description: "<taskSlug> QA"
  subagent_type: general-purpose
  team_name: "<TICKET>"
  name: "qa-task-<N>"
  mode: bypassPermissions
  model: haiku
  run_in_background: true
  prompt: <substitute THIN-SPAWN-TEMPLATE.md QA template>

Save the returned task_id for each agent.