workflows:swarm

Swarm-enabled autonomous workflow. Run these steps in order, parallelizing where indicated. Do not stop between steps - complete every step through to the end.

Steps 2, 5, 6 and 7 call /deepen-plan, /test-browser, /resolve_todo_parallel and /feature-video, which this kit does not ship. If a command is not installed, skip that step and say so in the run summary rather than stopping.

Dynamic Context

Current branch: !git branch --show-current Recent commits: !git log --oneline -5

Phase 1: Plan (Sequential)

1. /workflows:plan $ARGUMENTS
2. /deepen-plan

Phase 2: Execute (Swarm)

After the plan is written and deepened, execute it using parallel agent swarm orchestration.

Step 1: Create the Task Board

Read the plan file. Break it into discrete, parallelizable tasks. Create ALL tasks upfront using TaskCreate, then wire dependencies with TaskUpdate.

TaskCreate({
  subject: "Task 1: [specific deliverable]",
  description: "[what needs to be done, files to create/modify, references to follow]",
  activeForm: "Building [deliverable]..."
})
TaskCreate({
  subject: "Task 2: [specific deliverable]",
  description: "[what needs to be done]",
  activeForm: "Drafting [deliverable]..."
})
... (create every task from the plan)

Then set up dependencies:

TaskUpdate({ taskId: "3", addBlockedBy: ["1"] })  // Task 3 needs Task 1 done first
TaskUpdate({ taskId: "5", addBlockedBy: ["2", "4"] })  // Task 5 needs 2 and 4

Rules:

• Aim for 8-15 tasks. Each = a concrete deliverable (file, component, report, draft, config)
• Group tiny items together. Split large items apart.
• Independent tasks have NO blockedBy (can run in parallel immediately)
• Sequential tasks use addBlockedBy to enforce ordering
• The user sees real-time progress as dots change: pending -> in_progress -> completed

Step 2: Launch the Swarm

Launch 5-6 parallel Agent workers. Each agent gets a color for visual differentiation. Use isolation: "worktree" when agents touch different files to avoid conflicts.