name: team-perf-opt description: Unified team skill for performance optimization. Coordinator orchestrates pipeline, workers are team-worker agents. Supports single/fan-out/independent parallel modes. Triggers on "team perf-opt". allowed-tools: spawn_agent(), wait_agent(), send_message(), followup_task(), close_agent(), list_agents(), report_agent_job_result(), request_user_input(), Read(), Write(), Edit(), Bash(), Glob(), Grep(), mcp__ace-tool__search_context(), mcp__ccw-tools__team_msg()

Team Performance Optimization

Profile application performance, identify bottlenecks, design optimization strategies, implement changes, benchmark improvements, and review code quality.

Architecture

Skill(skill="team-perf-opt", 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 workers -> STOP
                                    |
                    +-------+-------+-------+-------+-------+
                    v       v       v       v       v
                 [profiler] [strategist] [optimizer] [benchmarker] [reviewer]
                 (team-worker agents)

Pipeline (Single mode):
  PROFILE-001 -> STRATEGY-001 -> IMPL-001 -> BENCH-001 + REVIEW-001 (fix cycle)

Pipeline (Fan-out mode):
  PROFILE-001 -> STRATEGY-001 -> [IMPL-B01..N](parallel) -> BENCH+REVIEW per branch

Pipeline (Independent mode):
  [Pipeline A: PROFILE-A->STRATEGY-A->IMPL-A->BENCH-A+REVIEW-A]
  [Pipeline B: PROFILE-B->STRATEGY-B->IMPL-B->BENCH-B+REVIEW-B] (parallel)

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: PERF-OPT
• Session path: .workflow/.team/PERF-OPT-<date>-<slug>/
• Team name: perf-opt
• 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|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.

Inner Loop roles (optimizer): Set inner_loop: true. Single-task roles (profiler, strategist, benchmarker, reviewer): Set inner_loop: false.

Model Selection Guide

Performance optimization is measurement-driven. Profiler and benchmarker need consistent context for before/after comparison.

Benchmark Context Sharing with fork_turns

For before/after comparison, benchmarker should share context with profiler's baseline:

spawn_agent({
  agent_type: "team_worker",
  task_name: "BENCH-001",
  fork_turns: "all",   // Share context so benchmarker sees profiler's baseline metrics
  reasoning_effort: "medium",
  message: "..."
})

User Commands

Session Directory

.workflow/.team/PERF-OPT-<date>-<slug>/
+-- session.json                    # Session metadata + status + parallel_mode
+-- artifacts/
|   +-- baseline-metrics.json       # Profiler: before-optimization metrics
|   +-- bottleneck-report.md        # Profiler: ranked bottleneck findings
|   +-- optimization-plan.md        # Strategist: prioritized optimization plan
|   +-- benchmark-results.json      # Benchmarker: after-optimization metrics
|   +-- review-report.md            # Reviewer: code review findings
|   +-- branches/B01/...            # Fan-out branch artifacts
|   +-- pipelines/A/...             # Independent pipeline artifacts
+-- explorations/                   # Shared explore cache
+-- wisdom/patterns.md              # Discovered patterns and conventions
+-- discussions/                    # Discussion records
+-- .msg/messages.jsonl             # Team message bus
+-- .msg/meta.json                  # Session metadata

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 session.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: "IMPL-001", message: "..." }) -- send strategy details to optimizer
• followup_task({ target: "IMPL-001", message: "..." }) -- assign fix after benchmark regression
• close_agent({ target: "BENCH-001" }) -- cleanup after benchmarking completes

Baseline-to-Result Pipeline

Profiler baseline metrics flow through the pipeline and must reach benchmarker for comparison:

1. PROFILE-001 produces baseline-metrics.json in artifacts/
2. Coordinator includes baseline reference in upstream context for all downstream workers
3. BENCH-001 reads baseline and compares against post-optimization measurements
4. If regression detected, coordinator auto-creates FIX task with regression details

Completion Action

When the pipeline completes:

functions.request_user_input({
  questions: [{
    question: "Team 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
• specs/team-config.json — Team configuration

Error Handling