team-planex-v2（平面团队 v2）

Auto Mode

When --yes or -y: Auto-confirm task decomposition, skip interactive validation, use defaults.

Team PlanEx

Usage

$team-planex-v2 "ISS-20260308-120000 ISS-20260308-120001"
$team-planex-v2 -c 3 "--text 'Add rate limiting to all API endpoints'"
$team-planex-v2 -y "--plan .workflow/specs/roadmap.md --exec=codex"
$team-planex-v2 --continue "planex-rate-limit-20260308"

Flags:

• -y, --yes: Skip all confirmations (auto mode)
• -c, --concurrency N: Max concurrent agents within each wave (default: 3)
• --continue: Resume existing session
• --exec=codex|gemini|qwen: Force execution method for implementation

Output Directory: .workflow/.csv-wave/{session-id}/ Core Output: tasks.csv (master state) + results.csv (final) + discoveries.ndjson (shared exploration) + context.md (human-readable report)

Overview

Plan-and-execute pipeline for issue-based development. Planner decomposes requirements into individual issues with solution plans, then executors implement each issue independently.

Execution Model: Hybrid -- CSV wave pipeline (primary) + individual agent spawn (secondary)

+---------------------------------------------------------------------------+
|                        TEAM PLANEX WORKFLOW                                |
+---------------------------------------------------------------------------+
|                                                                           |
|  Phase 0: Pre-Wave Interactive (Input Analysis)                           |
|     +-- Parse input type (issue IDs / --text / --plan)                    |
|     +-- Determine execution method (codex/gemini/auto)                    |
|     +-- Create issues from text/plan if needed                            |
|     +-- Output: refined issue list for decomposition                      |
|                                                                           |
|  Phase 1: Requirement -> CSV + Classification                             |
|     +-- Planning wave: generate solutions for each issue                  |
|     +-- Execution wave: implement each issue independently                |
|     +-- Classify tasks: csv-wave (default) | interactive                  |
|     +-- Compute dependency waves (topological sort)                       |
|     +-- Generate tasks.csv with wave + exec_mode columns                  |
|     +-- User validates task breakdown (skip if -y)                        |
|                                                                           |
|  Phase 2: Wave Execution Engine (Extended)                                |
|     +-- For each wave (1..N):                                             |
|     |   +-- Build wave CSV (filter csv-wave tasks for this wave)          |
|     |   +-- Inject previous findings into prev_context column             |
|     |   +-- spawn_agents_on_csv(wave CSV)                                 |
|     |   +-- Merge all results into master tasks.csv                       |
|     |   +-- Check: any failed? -> skip dependents                         |
|     +-- discoveries.ndjson shared across all modes (append-only)          |
|                                                                           |
|  Phase 3: Results Aggregation                                             |
|     +-- Export final results.csv                                          |
|     +-- Generate context.md with all findings                             |
|     +-- Display summary: completed/failed/skipped per wave                |
|     +-- Offer: view results | retry failed | done                         |
|                                                                           |
+---------------------------------------------------------------------------+

Task Classification Rules

Each task is classified by exec_mode:

Classification Decision:

In the standard PlanEx pipeline, all tasks default to csv-wave. Interactive mode is reserved for edge cases requiring multi-round coordination.

CSV Schema

tasks.csv (Master State)

id,title,description,role,issue_ids,input_type,raw_input,exec_mode,execution_method,deps,context_from,wave,status,findings,artifact_path,error
"PLAN-001","Plan issue-1","Generate solution for ISS-20260308-120000","planner","ISS-20260308-120000","issues","ISS-20260308-120000","csv-wave","","","","1","pending","","",""
"PLAN-002","Plan issue-2","Generate solution for ISS-20260308-120001","planner","ISS-20260308-120001","issues","ISS-20260308-120001","csv-wave","","","","1","pending","","",""
"EXEC-001","Implement issue-1","Implement solution for ISS-20260308-120000","executor","ISS-20260308-120000","","","csv-wave","gemini","PLAN-001","PLAN-001","2","pending","","",""
"EXEC-002","Implement issue-2","Implement solution for ISS-20260308-120001","executor","ISS-20260308-120001","","","csv-wave","gemini","PLAN-002","PLAN-002","2","pending","","",""

Columns:

Per-Wave CSV (Temporary)

Each wave generates a temporary wave-{N}.csv with extra prev_context column (csv-wave tasks only).

Output Artifacts

Session Structure

.workflow/.csv-wave/{session-id}/
+-- tasks.csv                  # Master state (all tasks)
+-- results.csv                # Final results export
+-- discoveries.ndjson         # Shared discovery board
+-- context.md                 # Human-readable report
+-- wave-{N}.csv               # Temporary per-wave input
+-- artifacts/
|   +-- solutions/             # Planner output
|       +-- {issueId}.json
+-- builds/                    # Executor output
|   +-- {issueId}.json
+-- wisdom/                    # Cross-task knowledge
    +-- learnings.md
    +-- decisions.md
    +-- conventions.md
    +-- issues.md

Implementation

Session Initialization

const getUtc8ISOString = () => new Date(Date.now() + 8 * 60 * 60 * 1000).toISOString()

const AUTO_YES = $ARGUMENTS.includes('--yes') || $ARGUMENTS.includes('-y')
const continueMode = $ARGUMENTS.includes('--continue')
const concurrencyMatch = $ARGUMENTS.match(/(?:--concurrency|-c)\s+(\d+)/)
const maxConcurrency = concurrencyMatch ? parseInt(concurrencyMatch[1]) : 3

const requirement = $ARGUMENTS
  .replace(/--yes|-y|--continue|--concurrency\s+\d+|-c\s+\d+/g, '')
  .trim()

// Parse execution method
let executionMethod = 'gemini' // default
const execMatch = requirement.match(/--exec=(\w+)/)
if (execMatch) executionMethod = execMatch[1]

// Detect input type
const issueIdPattern = /ISS-\d{8}-\d{6}/g
const textMatch = requirement.match(/--text\s+'([^']+)'/)
const planMatch = requirement.match(/--plan\s+(\S+)/)

let inputType = 'issues'
let rawInput = requirement
let issueIds = requirement.match(issueIdPattern) || []

if (textMatch) {
  inputType = 'text'
  rawInput = textMatch[1]
  issueIds = [] // will be created by planner
} else if (planMatch) {
  inputType = 'plan'
  rawInput = planMatch[1]
  issueIds = [] // will be parsed from plan file
}

// If no input detected, ask user
if (issueIds.length === 0 && inputType === 'issues') {
  const answer = AskUserQuestion("No input detected. Provide issue IDs, or use --text 'description' or --plan <path>:")
  issueIds = answer.match(issueIdPattern) || []
  if (issueIds.length === 0 && !answer.includes('--text') && !answer.includes('--plan')) {
    inputType = 'text'
    rawInput = answer
  }
}

// Execution method selection (interactive if no flag)
if (!execMatch && !AUTO_YES) {
  const methodChoice = AskUserQuestion({
    questions: [{ question: "Select execution method for implementation:",
      options: [
        { label: "Gemini", description: "gemini-2.5-pro (recommended for <= 3 tasks)" },
        { label: "Codex", description: "gpt-5.2 (recommended for > 3 tasks)" },
        { label: "Auto", description: "Auto-select based on task count" }
      ]
    }]
  })
  if (methodChoice === 'Codex') executionMethod = 'codex'
  else if (methodChoice === 'Auto') executionMethod = 'auto'
}

const slug = (issueIds[0] || rawInput).toLowerCase()
  .replace(/[^a-z0-9\u4e00-\u9fa5]+/g, '-')
  .substring(0, 30)
const dateStr = getUtc8ISOString().substring(0, 10).replace(/-/g, '')
const sessionId = `planex-${slug}-${dateStr}`
const sessionFolder = `.workflow/.csv-wave/${sessionId}`

Bash(`mkdir -p ${sessionFolder}/{artifacts/solutions,builds,wisdom}`)

Write(`${sessionFolder}/discoveries.ndjson`, `# Discovery Board - ${sessionId}\n# Format: NDJSON\n`)

// Initialize wisdom files
Write(`${sessionFolder}/wisdom/learnings.md`, `# Learnings\n\nAccumulated during ${sessionId}\n`)
Write(`${sessionFolder}/wisdom/decisions.md`, `# Decisions\n\n`)
Write(`${sessionFolder}/wisdom/conventions.md`, `# Conventions\n\n`)
Write(`${sessionFolder}/wisdom/issues.md`, `# Issues\n\n`)

// Store session metadata
Write(`${sessionFolder}/session.json`, JSON.stringify({
  session_id: sessionId,
  pipeline_type: 'plan-execute',
  input_type: inputType,
  raw_input: rawInput,
  issue_ids: issueIds,
  execution_method: executionMethod,
  created_at: getUtc8ISOString()
}, null, 2))

Phase 0: Pre-Wave Interactive (Input Analysis)

Objective: Parse and normalize input into a list of issue IDs ready for the planning wave.

Input Type Handling:

For text input:

# Create issue from text description
ccw issue create --title "<derived-title>" --context "<raw_input>"
# Parse output for new issue ID

For plan input:

# Read plan file
planContent = Read("<plan-path>")
# Parse phases/sections into individual issues
# Create each as a separate issue via ccw issue create

After processing, update session.json with resolved issue_ids.

Success Criteria:

• All inputs resolved to valid issue IDs
• Session metadata updated with final issue list

Phase 1: Requirement -> CSV + Classification

Objective: Generate tasks.csv with PLAN-* tasks (wave 1) and EXEC-* tasks (wave 2).

Two-Wave Structure:

Wave 1 (Planning): One PLAN-NNN task per issue, all independent (no deps), concurrent execution. Wave 2 (Execution): One EXEC-NNN task per issue, each depends on its corresponding PLAN-NNN.

Task Generation:

const tasks = []

// Wave 1: Planning tasks (one per issue)
for (let i = 0; i < issueIds.length; i++) {
  const n = String(i + 1).padStart(3, '0')
  tasks.push({
    id: `PLAN-${n}`,
    title: `Plan ${issueIds[i]}`,
    description: `Generate implementation solution for issue ${issueIds[i]}. Analyze requirements, design solution approach, break down into implementation tasks, identify files to modify/create.`,
    role: 'planner',
    issue_ids: issueIds[i],
    input_type: inputType,
    raw_input: inputType === 'issues' ? issueIds[i] : rawInput,
    exec_mode: 'csv-wave',
    execution_method: '',
    deps: '',
    context_from: '',
    wave: '1',
    status: 'pending',
    findings: '', artifact_path: '', error: ''
  })
}

// Wave 2: Execution tasks (one per issue, depends on corresponding PLAN)
for (let i = 0; i < issueIds.length; i++) {
  const n = String(i + 1).padStart(3, '0')
  // Resolve execution method
  let method = executionMethod
  if (method === 'auto') {
    method = issueIds.length <= 3 ? 'gemini' : 'codex'
  }
  tasks.push({
    id: `EXEC-${n}`,
    title: `Implement ${issueIds[i]}`,
    description: `Implement solution for issue ${issueIds[i]}. Load solution artifact, execute implementation via CLI, run tests, commit.`,
    role: 'executor',
    issue_ids: issueIds[i],
    input_type: '',
    raw_input: '',
    exec_mode: 'csv-wave',
    execution_method: method,
    deps: `PLAN-${n}`,
    context_from: `PLAN-${n}`,
    wave: '2',
    status: 'pending',
    findings: '', artifact_path: '', error: ''
  })
}

Write(`${sessionFolder}/tasks.csv`, toCsv(tasks))

User Validation: Display task breakdown with wave assignment (skip if AUTO_YES).

Success Criteria:

• tasks.csv created with valid schema and wave assignments
• PLAN-* tasks in wave 1, EXEC-* tasks in wave 2
• Each EXEC-* depends on its corresponding PLAN-*
• No circular dependencies
• User approved (or AUTO_YES)

Phase 2: Wave Execution Engine (Extended)

Objective: Execute tasks wave-by-wave with context propagation between planning and execution waves.

const masterCsv = Read(`${sessionFolder}/tasks.csv`)
let tasks = parseCsv(masterCsv)
const maxWave = Math.max(...tasks.map(t => parseInt(t.wave)))

for (let wave = 1; wave <= maxWave; wave++) {
  console.log(`\nWave ${wave}/${maxWave} (${wave === 1 ? 'Planning' : 'Execution'})`)

  // 1. Filter tasks for this wave
  const waveTasks = tasks.filter(t => parseInt(t.wave) === wave && t.status === 'pending')

  // 2. Check dependencies - skip if upstream failed
  for (const task of waveTasks) {
    const depIds = (task.deps || '').split(';').filter(Boolean)
    const depStatuses = depIds.map(id => tasks.find(t => t.id === id)?.status)
    if (depStatuses.some(s => s === 'failed' || s === 'skipped')) {
      task.status = 'skipped'
      task.error = `Dependency failed: ${depIds.filter((id, i) =>
        ['failed','skipped'].includes(depStatuses[i])).join(', ')}`
    }
  }

  const pendingTasks = waveTasks.filter(t => t.status === 'pending')
  if (pendingTasks.length === 0) {
    console.log(`Wave ${wave}: No pending tasks, skipping...`)
    continue
  }

  // 3. Build prev_context from completed upstream tasks
  for (const task of pendingTasks) {
    const contextIds = (task.context_from || '').split(';').filter(Boolean)
    const prevFindings = contextIds.map(id => {
      const src = tasks.find(t => t.id === id)
      if (!src?.findings) return ''
      return `## [${src.id}] ${src.title}\n${src.findings}\nArtifact: ${src.artifact_path || 'N/A'}`
    }).filter(Boolean).join('\n\n')
    task.prev_context = prevFindings
  }

  // 4. Write wave CSV
  Write(`${sessionFolder}/wave-${wave}.csv`, toCsv(pendingTasks))

  // 5. Execute wave
  spawn_agents_on_csv({
    csv_path: `${sessionFolder}/wave-${wave}.csv`,
    id_column: "id",
    instruction: Read("~  or <project>/.codex/skills/team-planex/instructions/agent-instruction.md"),
    max_concurrency: maxConcurrency,
    max_runtime_seconds: 1200,
    output_csv_path: `${sessionFolder}/wave-${wave}-results.csv`,
    output_schema: {
      type: "object",
      properties: {
        id: { type: "string" },
        status: { type: "string", enum: ["completed", "failed"] },
        findings: { type: "string" },
        artifact_path: { type: "string" },
        error: { type: "string" }
      }
    }
  })

  // 6. Merge results into master CSV
  const results = parseCsv(Read(`${sessionFolder}/wave-${wave}-results.csv`))
  for (const r of results) {
    const t = tasks.find(t => t.id === r.id)
    if (t) Object.assign(t, r)
  }
  Write(`${sessionFolder}/tasks.csv`, toCsv(tasks))

  // 7. Cleanup temp files
  Bash(`rm -f ${sessionFolder}/wave-${wave}.csv ${sessionFolder}/wave-${wave}-results.csv`)

  // 8. Display wave summary
  const completed = results.filter(r => r.status === 'completed').length
  const failed = results.filter(r => r.status === 'failed').length
  console.log(`Wave ${wave} Complete: ${completed} completed, ${failed} failed`)
}

Success Criteria:

• All waves executed in order
• Each wave's results merged into master CSV before next wave starts
• Dependent tasks skipped when predecessor failed
• discoveries.ndjson accumulated across all waves
• Planning wave completes before execution wave starts

Phase 3: Results Aggregation

Objective: Generate final results and human-readable report.

const tasks = parseCsv(Read(`${sessionFolder}/tasks.csv`))
const completed = tasks.filter(t => t.status === 'completed')
const failed = tasks.filter(t => t.status === 'failed')
const skipped = tasks.filter(t => t.status === 'skipped')

const planTasks = tasks.filter(t => t.role === 'planner')
const execTasks = tasks.filter(t => t.role === 'executor')

// Export results.csv
Bash(`cp ${sessionFolder}/tasks.csv ${sessionFolder}/results.csv`)

// Generate context.md
let contextMd = `# PlanEx Pipeline Report\n\n`
contextMd += `**Session**: ${sessionId}\n`
contextMd += `**Input Type**: ${inputType}\n`
contextMd += `**Execution Method**: ${executionMethod}\n`
contextMd += `**Issues**: ${issueIds.join(', ')}\n\n`

contextMd += `## Summary\n\n`
contextMd += `| Status | Count |\n|--------|-------|\n`
contextMd += `| Completed | ${completed.length} |\n`
contextMd += `| Failed | ${failed.length} |\n`
contextMd += `| Skipped | ${skipped.length} |\n\n`

contextMd += `## Planning Wave\n\n`
for (const t of planTasks) {
  const icon = t.status === 'completed' ? '[OK]' : t.status === 'failed' ? '[FAIL]' : '[SKIP]'
  contextMd += `${icon} **${t.id}**: ${t.title}\n`
  if (t.findings) contextMd += `   ${t.findings.substring(0, 200)}\n`
  if (t.artifact_path) contextMd += `   Solution: ${t.artifact_path}\n`
  contextMd += `\n`
}

contextMd += `## Execution Wave\n\n`
for (const t of execTasks) {
  const icon = t.status === 'completed' ? '[OK]' : t.status === 'failed' ? '[FAIL]' : '[SKIP]'
  contextMd += `${icon} **${t.id}**: ${t.title}\n`
  if (t.findings) contextMd += `   ${t.findings.substring(0, 200)}\n`
  if (t.error) contextMd += `   Error: ${t.error}\n`
  contextMd += `\n`
}

contextMd += `## Deliverables\n\n`
contextMd += `| Artifact | Path |\n|----------|------|\n`
contextMd += `| Solution Plans | ${sessionFolder}/artifacts/solutions/ |\n`
contextMd += `| Build Results | ${sessionFolder}/builds/ |\n`
contextMd += `| Discovery Board | ${sessionFolder}/discoveries.ndjson |\n`

Write(`${sessionFolder}/context.md`, contextMd)

// Display summary
console.log(`
PlanEx Pipeline Complete
Input: ${inputType} (${issueIds.length} issues)
Planning: ${planTasks.filter(t => t.status === 'completed').length}/${planTasks.length} completed
Execution: ${execTasks.filter(t => t.status === 'completed').length}/${execTasks.length} completed
Failed: ${failed.length} | Skipped: ${skipped.length}
Output: ${sessionFolder}
`)

Success Criteria:

• results.csv exported (all tasks)
• context.md generated
• Summary displayed to user

Shared Discovery Board Protocol

Both planner and executor agents share the same discoveries.ndjson file:

{"ts":"2026-03-08T10:00:00Z","worker":"PLAN-001","type":"solution_designed","data":{"issue_id":"ISS-20260308-120000","approach":"refactor","task_count":4,"estimated_files":6}}
{"ts":"2026-03-08T10:05:00Z","worker":"PLAN-002","type":"conflict_warning","data":{"issue_ids":["ISS-20260308-120000","ISS-20260308-120001"],"overlapping_files":["src/auth/handler.ts"]}}
{"ts":"2026-03-08T10:10:00Z","worker":"EXEC-001","type":"impl_result","data":{"issue_id":"ISS-20260308-120000","files_changed":3,"tests_pass":true,"commit":"abc123"}}

Discovery Types:

Error Handling

Core Rules

1. Start Immediately: First action is session initialization, then input parsing
2. Wave Order is Sacred: Never execute wave N before wave N-1 completes and results are merged
3. CSV is Source of Truth: Master tasks.csv holds all state
4. CSV First: Default to csv-wave for all tasks; interactive only for edge cases
5. Context Propagation: prev_context built from master CSV, not from memory
6. Discovery Board is Append-Only: Never clear, modify, or recreate discoveries.ndjson
7. Skip on Failure: If PLAN-N failed, skip EXEC-N automatically
8. Cleanup Temp Files: Remove wave-{N}.csv after results are merged
9. Two-Wave Pipeline: Wave 1 = Planning (PLAN-*), Wave 2 = Execution (EXEC-*)
10. DO NOT STOP: Continuous execution until all waves complete or all remaining tasks are skipped

Coordinator Role Constraints (Main Agent)

CRITICAL: The coordinator (main agent executing this skill) is responsible for orchestration only, NOT implementation.

1. Coordinator Does NOT Execute Code: The main agent MUST NOT write, modify, or implement any code directly. All implementation work is delegated to spawned team agents. The coordinator only:

- Spawns agents with task assignments - Waits for agent callbacks - Merges results and coordinates workflow - Manages workflow transitions between phases

1. Patient Waiting is Mandatory: Agent execution takes significant time (typically 10-30 minutes per phase, sometimes longer). The coordinator MUST:

- Wait patiently for wait() calls to complete - NOT skip workflow steps due to perceived delays - NOT assume agents have failed just because they're taking time - Trust the timeout mechanisms defined in the skill

1. Use send_input for Clarification: When agents need guidance or appear stuck, the coordinator MUST:

- Use send_input() to ask questions or provide clarification - NOT skip the agent or move to next phase prematurely - Give agents opportunity to respond before escalating - Example: send_input({id: agent_id, message: "Please provide status update or clarify blockers"})

1. No Workflow Shortcuts: The coordinator MUST NOT:

- Skip phases or stages defined in the workflow - Bypass required approval or review steps - Execute dependent tasks before prerequisites complete - Assume task completion without explicit agent callback - Make up or fabricate agent results

1. Respect Long-Running Processes: This is a complex multi-agent workflow that requires patience:

- Total execution time may range from 30-90 minutes or longer - Each phase may take 10-30 minutes depending on complexity - The coordinator must remain active and attentive throughout the entire process - Do not terminate or skip steps due to time concerns