delegate

name	delegate
description	Autonomous execution — classify, allocate, and execute without confirmation

When this skill is invoked, you MUST follow this exact output sequence. Do NOT ask for confirmation at any point.

Step 1: Header (print IMMEDIATELY)

[BW] delegate engaged — executing autonomously...

Step 2: Classify and allocate (silent, no user prompt)

Analyze the user's task:

1. Split into sub-tasks (1-5)
2. Detect domains per sub-task
3. Assign agents per sub-task (tech only — no PM/critic overhead in delegate)
4. Pick the fastest execution path

Print the plan inline (do NOT wait for confirmation):

[BW] delegate plan:
  1. "{task}" → bestwork:{agent}
  2. "{task}" → bestwork:{agent}
[BW] executing...

Step 3: Execute immediately

For single task:

• Load the specialist's prompt from prompts/{role}/{name}.md
• Execute directly as that specialist

For multiple tasks:

• Spawn all agents in parallel using run_in_background: true
• For EACH agent, call TaskCreate for live progress spinner: TaskCreate: subject="bestwork:{agent} (task N)", activeForm="bestwork:{agent} working..."
• Print each launch:

[BW] ▶ delegate: bestwork:{agent} (task 1)
[BW] ▶ delegate: bestwork:{agent} (task 2)

Step 4: Results (print as each completes, TaskUpdate to completed)

[BW] ✓ bestwork:{agent} done (task 1) — {summary}
[BW] ✓ bestwork:{agent} done (task 2) — {summary}

Step 5: Self-verify (quick sanity check, no critic agent)

After all tasks complete, do a quick self-check:

1. If code was written: run npx tsc --noEmit or equivalent type check
2. If tests exist: run the project's test command
3. If files were created: verify they are importable

Print result:

[BW] delegate verify: {PASS|FAIL}

If FAIL: fix the issue immediately (1 retry max), then move on.

Step 6: Summary

[BW] ═══════════════════════════════════
[BW] delegate complete: {N} tasks, {M} agents
[BW] ═══════════════════════════════════

Rules

• NEVER ask for confirmation — that is the entire point of delegate
• NEVER spawn PM or critic agents — delegate is speed-optimized
• Tech agents only — pick the best-fit specialist per task
• Self-verify replaces critic review (type check + test run)
• Max 1 self-fix retry on verify failure
• Do NOT write meeting logs — delegate skips ceremony
• If the task is ambiguous, make a reasonable choice and proceed (do not ask)