Agent Skills: Team Roadmap Dev

Team Roadmap Dev

Roadmap-driven development with phased execution pipeline. Coordinator discusses roadmap with the user and manages phase transitions. Workers are spawned as team-worker agents.

Architecture

Skill(skill="team-roadmap-dev", args="<task-description>")
                    |
         SKILL.md (this file) = Router
                    |
     +--------------+--------------+
     |                             |
  no --role flag              --role <name>
     |                             |
  Coordinator                  Worker
  roles/coordinator/role.md    roles/<name>/role.md
     |
     +-- roadmap-discuss -> dispatch -> spawn workers -> STOP
                                    |
                    +-------+-------+-------+
                    v       v       v
                 [planner] [executor] [verifier]
                 (team-worker agents)

Pipeline (per phase):
  PLAN-N01 -> EXEC-N01 -> VERIFY-N01 (gap closure loop if needed)

Multi-phase:
  Phase 1 -> Phase 2 -> ... -> Phase N -> Complete

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: RD
• Session path: .workflow/.team/RD-<date>-<slug>/
• Team name: roadmap-dev
• CLI tools: ccw cli --mode analysis (read-only), ccw cli --mode write (modifications)
• Message bus: mcp__ccw-tools__team_msg(session_id=<session-id>, ...)

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

Read role_spec file (<skill_root>/roles/<role>/role.md) to load Phase 2-4 domain instructions.

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

## Upstream Context
<prev_context>`
})

After spawning, use wait_agent({ timeout_ms: 1800000 }) to collect results. If result.timed_out, send STATUS_CHECK via followup_task (wait 3 min), then FINALIZE with interrupt (wait 3 min), then mark timed_out and close agents. Use close_agent({ target }) each worker.

All worker roles (planner, executor, verifier): Set inner_loop: true.

Model Selection Guide

Roadmap development is context-heavy with multi-phase execution. All roles use inner_loop and need high reasoning for complex planning/execution.

All roles are inner_loop=true, enabling coordinator to send additional context via followup_task as phases progress.

User Commands

Session Directory

.workflow/.team/RD-<date>-<slug>/
+-- roadmap.md                 # Phase plan with requirements and success criteria
+-- state.md                   # Living memory (<100 lines)
+-- config.json                # Session settings (mode, depth, gates)
+-- wisdom/                    # Cross-task knowledge accumulation
|   +-- learnings.md
|   +-- decisions.md
|   +-- conventions.md
|   +-- issues.md
+-- phase-1/                   # Per-phase artifacts
|   +-- context.md
|   +-- IMPL_PLAN.md
|   +-- TODO_LIST.md
|   +-- .task/IMPL-*.json
|   +-- summary-*.md
|   +-- verification.md
+-- phase-N/
|   +-- ...
+-- .msg/
    +-- messages.jsonl          # Team message bus log
    +-- meta.json               # Session metadata + shared state

v4 Agent Coordination

Message Semantics

Agent Health Check

Use list_agents({}) in handleResume and handleComplete:

// Reconcile session state with actual running agents
const running = list_agents({})
// Compare with state.md and config.json active tasks
// Reset orphaned tasks (in_progress but agent gone) to pending

Named Agent Targeting

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

• send_message({ target: "EXEC-N01", message: "..." }) -- send supplementary context to executor
• followup_task({ target: "PLAN-N01", message: "..." }) -- assign next phase planning
• close_agent({ target: "VERIFY-N01" }) -- cleanup after verification

Multi-Phase Context Accumulation

Each phase builds on previous phase results. Coordinator accumulates context across phases:

• Phase N planner receives: roadmap.md + state.md + all previous phase summaries
• Phase N executor receives: phase plan + previous phase implementation context
• Phase N verifier receives: phase plan + executor results + success criteria from roadmap
• On gap closure: verifier findings are sent back to executor via followup_task (max 3 iterations)

Completion Action

When the pipeline completes:

functions.request_user_input({
  questions: [{
    question: "Roadmap Dev pipeline complete. What would you like to do?",
    header: "Completion",
    multiSelect: false,
    options: [
      { label: "Archive & Clean (Recommended)", description: "Archive session, clean up tasks and team resources" },
      { label: "Keep Active", description: "Keep session active for follow-up work or inspection" },
      { label: "Export Results", description: "Export deliverables to a specified location, then clean" }
    ]
  }]
})

Specs Reference

• specs/pipelines.md — Pipeline definitions and task registry

Error Handling