Agent Skills: Team Issue Resolution

Team Issue Resolution

Orchestrate issue resolution pipeline: explore context -> plan solution -> review (optional) -> marshal queue -> implement. Supports Quick, Full, and Batch pipelines with review-fix cycle.

Architecture

Skill(skill="team-issue", args="<issue-ids> [--mode=<mode>]")
                    |
         SKILL.md (this file) = Router
                    |
     +--------------+--------------+
     |                             |
  no --role flag              --role <name>
     |                             |
  Coordinator                  Worker
  roles/coordinator/role.md    roles/<name>/role.md
     |
     +-- clarify -> dispatch -> spawn workers -> STOP
                                    |
             +-------+-------+-------+-------+
             v       v       v       v       v
          [explor] [plann] [review] [integ] [imple]

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: TISL
• Session path: .workflow/.team/TISL-<date>-<slug>/
• Team name: issue
• 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: false

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.

Parallel spawn (Batch mode, N explorer or M implementer instances):

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>
agent_name: <role>-<N>
inner_loop: false

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.

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: "..."
})

User Commands

Session Directory

.workflow/.team/TISL-<date>-<slug>/
├── session.json                    # Session metadata + pipeline + fix_cycles
├── task-analysis.json              # Coordinator analyze output
├── .msg/
│   ├── messages.jsonl              # Message bus log
│   └── meta.json                   # Session state + cross-role state
├── wisdom/                         # Cross-task knowledge
│   ├── learnings.md
│   ├── decisions.md
│   ├── conventions.md
│   └── issues.md
├── explorations/                   # Explorer output
│   └── context-<issueId>.json
├── solutions/                      # Planner output
│   └── solution-<issueId>.json
├── audits/                         # Reviewer output
│   └── audit-report.json
├── queue/                          # Integrator output (also .workflow/issues/queue/)
└── builds/                         # Implementer output

Specs Reference

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

v4 Agent Coordination

Message Semantics

Pipeline Pattern

Pipeline with context passing: explore -> plan -> review (optional) -> marshal -> implement. In Batch mode, N explorers and M implementers run in parallel:

// Batch mode: spawn N explorers in parallel (max 5)
const explorerNames = ["EXPLORE-001", "EXPLORE-002", ..., "EXPLORE-00N"]
for (const name of explorerNames) {
  spawn_agent({ agent_type: "team_worker", task_name: name, ... })
}
wait_agent({ timeout_ms: 1800000 })  // 30 min — apply timeout cascade if timed_out

// After MARSHAL completes: spawn M implementers in parallel (max 3)
const buildNames = ["BUILD-001", "BUILD-002", ..., "BUILD-00M"]
for (const name of buildNames) {
  spawn_agent({ agent_type: "team_worker", task_name: name, ... })
}
wait_agent({ timeout_ms: 1800000 })  // 30 min — apply timeout cascade if timed_out

Review-Fix Cycle

Reviewer (AUDIT-*) may reject plans, triggering fix cycles (max 2). Dynamic SOLVE-fix and AUDIT re-review tasks are created in tasks.json.

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

Named Agent Targeting

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

• send_message({ target: "SOLVE-001", message: "..." }) -- queue exploration context to running planner
• close_agent({ target: "BUILD-001" }) -- cleanup by name after completion

Error Handling