PM Orchestrator (Lite)

Note: This version does not use fork subagents. All agents run directly in the current session.

Role

Orchestrator that sequentially executes PM analysis skills and constructs the final agent chain.

Input

Automatically collects the following:

• userMessage: User request
• gitBranch: Current branch
• gitStatus: Git status (clean/dirty)
• recentCommits: Recent commit list
• openFiles: Open file list

Workflow

1. Initialize analysisContext

schemaVersion: "1.0"
request:
  userMessage: "{userMessage}"
  taskType: unknown
  keywords: []
repo:
  gitBranch: "{gitBranch}"
  gitStatus: "{gitStatus}"
  openFiles: []
  changedFiles: []
signals:
  hasContextMd: false
  hasPendingQuestions: false
  requirementsClear: false
  implementationReady: false
  implementationComplete: false
  hasMockImplementation: false
  apiSpecConfirmed: false
  reactProject: false
estimates:
  estimatedFiles: 0
  estimatedLines: 0
  estimatedTime: unknown
phase: unknown
complexity: unknown
missingInfo: []
decisions:
  recommendedAgents: []
  skillChain: []
  parallelGroups: []
artifacts:
  tasksRoot: "{PROJECT.md:documentPaths.tasksRoot}"
  contextDocPath: "{tasksRoot}/{feature-name}/context.md"
  verificationScript: .claude/agents/verification/verify-changes.sh
tokenBudget:
  specSummaryTrigger: 2000
  splitTrigger: 5
  contextMaxTokens: 8000
  warningThreshold: 0.8
projectMemory:
  projectId: null
  loaded: false
  boundaries: null
  relevantRules: []
notes: []

2. Sequential PM Skill Execution

2.0 Large Specification Handling

2.0.1 Check Specification Size

• Count words in userMessage
• Trigger summarization if > tokenBudget.specSummaryTrigger (2000 words)
• Trigger task splitting if independent features > tokenBudget.splitTrigger (5)

2.0.2 Summarize Specification

1. Save original to {tasksRoot}/{feature-name}/archives/specification-full.md
2. Extract core elements only
3. Write summary to {tasksRoot}/{feature-name}/specification.md

2.0.3 Split into Subtasks If single spec covers multiple independent areas, split into subtasks/ directory.

2.0.4 Load Project Memory (Direct Execution)

Determine Project ID: package.json name → directory name → git remote

Direct Execution (no fork):

Task tool: project-memory-agent (subagent_type: general-purpose)
Input: { projectId, changedFiles, taskType, userRequest }

Agent searches project memory ([ProjectID]::*) and returns context:

projectMemoryContext:
  projectId: "my-app"
  loaded: true
  boundaries:
    alwaysDo: [...]
    askFirst: [...]
    neverDo: [...]
  relevantRules: [...]

Error Handling:

• No memory: loaded: false, continue
• MCP unavailable: continue with warning

2.1 Task Classification

Run /moonshot-classify-task using Skill tool

2.2 Complexity Evaluation

Run /moonshot-evaluate-complexity using Skill tool

2.3 Uncertainty Detection

Run /moonshot-detect-uncertainty using Skill tool

2.4 Handle Uncertainties

If missingInfo is not empty, generate questions using AskUserQuestion tool.

2.5 Sequence Decision

Run /moonshot-decide-sequence using Skill tool

3. Execute Agent Chain

Execute decisions.skillChain in order:

Allowed Steps:

• pre-flight-check: Pre-flight check skill
• project-memory-agent: Project memory load agent (Task tool, direct execution)
• requirements-analyzer: Requirements analysis agent (Task tool)
• context-builder: Context building agent (Task tool)
• codex-validate-plan: Codex plan validation skill
• implementation-runner: Implementation agent (Task tool)
• completion-verifier: Test-based completion verification skill
• codex-review-code: Codex code review skill
• project-memory-reviewer: Project memory rule/spec violation verification agent (Task tool, direct execution)
• vercel-react-best-practices: React/Next.js performance optimization review skill
• security-reviewer: Security vulnerability review skill
• build-error-resolver: Build/compile error resolution skill
• verify-changes.sh: Verification script (Bash tool)
• efficiency-tracker: Efficiency tracking skill
• session-logger: Session logging skill

Execution Rules:

1. Execute each step sequentially
2. Use Skill tool for skill steps
3. Use Task tool for agent steps (subagent_type mapping)
4. Use Bash tool for script steps
5. Parallel execution only within parallel groups
6. Request user confirmation for undefined steps
7. All agents/skills must follow .claude/docs/guidelines/document-memory-policy.md

Agent Mapping:

• project-memory-agent → subagent_type: "general-purpose" + prompt (direct execution)
• requirements-analyzer → subagent_type: "general-purpose" + prompt
• context-builder → subagent_type: "context-builder"
• implementation-runner → subagent_type: "implementation-agent"
• project-memory-reviewer → subagent_type: "general-purpose" + prompt (direct execution)

3.1 Dynamic Skill Injection

Inject skills dynamically when signals are detected during skillChain execution:

3.2 Project Memory Review (Direct Execution)

After codex-review-code, directly execute project-memory-reviewer:

Task tool: project-memory-reviewer (subagent_type: general-purpose)
Input: { projectId, changedFiles, projectMemoryContext, diff }

Receive Violation Report:

memoryReviewResult:
  status: "passed" | "failed" | "needs_approval"
  violations: [...]     # NeverDo violations
  needsApproval: [...]  # AskFirst items
  warnings: [...]       # Convention/spec warnings
  reminders: [...]      # AlwaysDo reminders

Result Handling:

• status: "failed": Stop execution, report violations to user
• status: "needs_approval": Request user approval before proceeding
• status: "passed": Proceed to next step

3.3 Completion Verification Loop

After implementation-runner completes, call completion-verifier. Retry up to retryCount < 2 if allPassed: false.

4. Record Results

Save final analysisContext to .claude/docs/moonshot-analysis.yaml.

Output Format

Summary for User (Markdown)

## PM Analysis Results

**Task Type**: {taskType}
**Complexity**: {complexity}
**Phase**: {phase}

### Execution Chain
1. {step1}
2. {step2}
...

### Estimates
- Files: {estimatedFiles}
- Lines: {estimatedLines}
- Estimated Time: {estimatedTime}

Error Handling

1. Skill execution failure: Log error to notes and report to user
2. Undefined step: Request user confirmation
3. Question infinite loop: Max 3 questions limit
4. Token limit warning: Archive and summarize before continuing

Contract

• This skill only orchestrates other PM skills, does not analyze directly
• All analysis logic delegated to individual PM skills
• Does not use fork - all agents run directly in current session
• Document memory policy: Follow .claude/docs/guidelines/document-memory-policy.md

References

• .claude/docs/guidelines/document-memory-policy.md