Team Lifecycle V4

Team Lifecycle v4

Orchestrate multi-agent software development: specification -> planning -> implementation -> testing -> review.

Architecture

Skill(skill="team-lifecycle-v4", args="task description")
                    |
         SKILL.md (this file) = Router
                    |
     +--------------+--------------+
     |                             |
  no --role flag              --role <name>
     |                             |
  Coordinator                  Worker
  roles/coordinator/role.md    roles/<name>/role.md
     |
     +-- analyze -> dispatch -> spawn -> wait -> collect
                                 |
                    +--------+---+--------+
                    v        v            v
            spawn_agent    ...     spawn_agent
          (team_worker)         (team_supervisor)
              per-task             resident agent
              lifecycle            followup_task-driven
                    |                     |
                    +-- wait_agent --------+
                              |
                         collect results

Role Registry

Role Router

Parse $ARGUMENTS:

• Has --role <name> -> Read roles/<name>/role.md, execute Phase 2-4
• No --role -> roles/coordinator/role.md, execute entry router

Delegation Lock

Coordinator is a PURE ORCHESTRATOR. It coordinates, it does NOT do.

Before calling ANY tool, apply this check:

If a tool call is BLOCKED: STOP. Create a task, spawn a worker.

No exceptions for "simple" tasks. Even a single-file read-and-report MUST go through spawn_agent.

Shared Constants

• Session prefix: TLV4
• Session path: .workflow/.team/TLV4-<date>-<slug>/
• State file: <session>/tasks.json
• Discovery files: <session>/discoveries/{task_id}.json
• CLI tools: ccw cli --mode analysis (read-only), ccw cli --mode write (modifications)

Worker Spawn Template

Coordinator spawns workers using this template:

spawn_agent({
  agent_type: "team_worker",
  task_name: "<task-id>",
  fork_turns: "none",
  message: `## Role Assignment
role: <role>
role_spec: <skill_root>/roles/<role>/role.md
session: <session-folder>
session_id: <session-id>
requirement: <task-description>
inner_loop: <true|false>

Read role_spec file (<skill_root>/roles/<role>/role.md) to load Phase 2-4 domain instructions.
Execute built-in Phase 1 (task discovery) -> role Phase 2-4 -> built-in Phase 5 (report).

## Task Context
task_id: <task-id>
title: <task-title>
description: <task-description>
pipeline_phase: <pipeline-phase>

## Upstream Context
<prev_context>`
})

Supervisor Spawn Template

Supervisor is a resident agent (independent from team_worker). Spawned once during session init, woken via followup_task for each CHECKPOINT task.

Spawn (Phase 2 -- once per session)

supervisorId = spawn_agent({
  agent_type: "team_supervisor",
  task_name: "supervisor",
  fork_turns: "none",
  message: `## Role Assignment
role: supervisor
role_spec: <skill_root>/roles/supervisor/role.md
session: <session-folder>
session_id: <session-id>
requirement: <task-description>

Read role_spec file (<skill_root>/roles/supervisor/role.md) to load checkpoint definitions.
Init: load baseline context, report ready, go idle.
Wake cycle: orchestrator sends checkpoint requests via followup_task.`
})

Wake (per CHECKPOINT task)

followup_task({
  target: "supervisor",
  message: `## Checkpoint Request
task_id: <CHECKPOINT-NNN>
scope: [<upstream-task-ids>]
pipeline_progress: <done>/<total> tasks completed`
})
wait_agent({ timeout_ms: 1800000 })  // 30 min — apply timeout cascade if timed_out

Shutdown (pipeline complete)

close_agent({ target: "supervisor" })

Model Selection Guide

Override model/reasoning_effort in spawn_agent when cost optimization is needed:

spawn_agent({
  agent_type: "team_worker",
  task_name: "<task-id>",
  fork_turns: "none",
  model: "<model-override>",
  reasoning_effort: "<effort-level>",
  message: "..."
})

Wave Execution Engine

For each wave in the pipeline:

1. Load state -- Read <session>/tasks.json, filter tasks for current wave
2. Skip failed deps -- Mark tasks whose dependencies failed/skipped as skipped
3. Build upstream context -- For each task, gather findings from context_from tasks via tasks.json and discoveries/{id}.json
4. Separate task types -- Split into regular tasks and CHECKPOINT tasks
5. Spawn regular tasks -- For each regular task, call spawn_agent({ agent_type: "team_worker", message: "..." }), collect agent IDs
6. Wait -- wait_agent({ timeout_ms: 1800000 }) — apply timeout cascade if timed_out
7. Collect results -- Read discoveries/{task_id}.json for each agent, update tasks.json status/findings/error, then close_agent({ target }) each worker
8. Execute checkpoints -- For each CHECKPOINT task, followup_task to supervisor, wait_agent, read checkpoint report from artifacts/, parse verdict
9. Handle block -- If verdict is block, prompt user via request_user_input with options: Override / Revise upstream / Abort
10. Persist -- Write updated state to <session>/tasks.json

User Commands

v4 Agent Coordination

Message Semantics

CRITICAL: The supervisor is a resident agent woken via followup_task, NOT send_message. Regular workers complete and are closed; the supervisor persists across checkpoints. See "Supervisor Spawn Template" above.

Agent Health Check

Use list_agents({}) in handleResume and handleComplete:

// Reconcile session state with actual running agents
const running = list_agents({})
// Compare with tasks.json active_agents
// Reset orphaned tasks (in_progress but agent gone) to pending
// ALSO check supervisor: if supervisor missing but CHECKPOINT tasks pending -> respawn

Named Agent Targeting

Workers are spawned with task_name: "<task-id>" enabling direct addressing:

• send_message({ target: "IMPL-001", message: "..." }) -- queue planning context to running implementer
• followup_task({ target: "supervisor", message: "..." }) -- wake supervisor for checkpoint
• close_agent({ target: "IMPL-001" }) -- cleanup regular worker by name
• close_agent({ target: "supervisor" }) -- shutdown supervisor at pipeline end

Completion Action

When pipeline completes, coordinator presents:

functions.request_user_input({
  questions: [{
    question: "Pipeline complete. What would you like to do?",
    header: "Completion",
    multiSelect: false,
    options: [
      { label: "Archive & Clean (Recommended)", description: "Archive session, clean up resources" },
      { label: "Keep Active", description: "Keep session for follow-up work" },
      { label: "Export Results", description: "Export deliverables to target directory" }
    ]
  }]
})

Specs Reference

• specs/pipelines.md -- Pipeline definitions and task registry
• specs/quality-gates.md -- Quality gate criteria and scoring
• specs/knowledge-transfer.md -- Artifact and state transfer protocols

Session Directory

.workflow/.team/TLV4-<date>-<slug>/
├── tasks.json                  # Task state (JSON)
├── discoveries/                # Per-task findings ({task_id}.json)
├── spec/                       # Spec phase outputs
├── plan/                       # Implementation plan
├── artifacts/                  # All deliverables
├── wisdom/                     # Cross-task knowledge
├── explorations/               # Shared explore cache
└── discussions/                # Discuss round records

Error Handling