Skip to content

agents-guide.md

Latest commit

 

History

History
306 lines (243 loc) · 6.83 KB

agents-guide.md

File metadata and controls

306 lines (243 loc) · 6.83 KB
Raw

日本語版

Agents Guide

A detailed guide to the sub-agents available in CommandMate.

What are Agents?

Agents are sub-processes that operate as specialists for specific tasks. They are invoked by orchestration commands such as PM Auto-Dev (/pm-auto-dev) and PM Bug Fix (/bug-fix), receive context files as input, execute processing, and output result files.

Agent List

Agent Description Called From
tdd-impl-agent TDD implementation specialist /tdd-impl, /pm-auto-dev
progress-report-agent Progress report generation /progress-report, /pm-auto-dev
investigation-agent Bug investigation specialist /bug-fix
acceptance-test-agent Acceptance testing /acceptance-test, /pm-auto-dev
refactoring-agent Refactoring /refactoring, /pm-auto-dev

tdd-impl-agent

Overview

A specialist agent that implements high-quality code following the TDD (Test-Driven Development) methodology.

Input

Context file: tdd-context.json

{
  "issue_number": 166,
  "acceptance_criteria": ["Criterion 1", "Criterion 2"],
  "implementation_tasks": ["Task 1", "Task 2"],
  "target_coverage": 80
}

Output

Result file: tdd-result.json

{
  "status": "success",
  "coverage": 85.0,
  "unit_tests": {
    "total": 10,
    "passed": 10,
    "failed": 0
  },
  "static_analysis": {
    "eslint_errors": 0,
    "typescript_errors": 0
  },
  "files_changed": ["src/lib/xxx.ts"],
  "commits": ["abc1234: feat(xxx): implement feature"]
}

Completion Criteria

  • All tests pass
  • Coverage meets or exceeds target
  • Zero static analysis errors
  • Commits complete

progress-report-agent

Overview

A specialist agent that aggregates results from each development phase and creates a progress report.

Input

Context file: progress-context.json

{
  "issue_number": 166,
  "iteration": 1,
  "phase_results": {
    "tdd": { "status": "success", "coverage": 85.0 },
    "acceptance": { "status": "passed" },
    "refactor": { "status": "success" }
  }
}

Output

Result file: progress-report.md

A progress report in Markdown format.

Completion Criteria

  • All result files loaded
  • Git history reviewed
  • Quality metrics aggregated
  • Next steps proposed

investigation-agent

Overview

A specialist agent for understanding the current state of bugs and investigating their causes. It identifies root causes through error log analysis, code investigation, and dependency checking, then presents recommended solutions.

Input

Context file: investigation-context.json

{
  "issue_description": "Error description",
  "error_logs": ["Error log 1"],
  "affected_files": ["src/lib/xxx.ts"],
  "reproduction_steps": ["1. xxx", "2. yyy"],
  "environment": {
    "os": "macOS",
    "node_version": "18.x"
  },
  "severity_hint": "high"
}

Output

Result file: investigation-result.json

{
  "status": "completed",
  "root_cause_analysis": {
    "category": "Code bug",
    "primary_cause": "Unhandled null reference"
  },
  "recommended_actions": [
    {
      "action_id": "1",
      "priority": "high",
      "title": "Solution 1",
      "description": "Solution description"
    }
  ]
}

Completion Criteria

  • Root cause identified
  • At least one solution proposed
  • Priority and risk assessed

acceptance-test-agent

Overview

A specialist agent that automatically runs acceptance tests based on Issue requirements and verifies that all acceptance criteria are met.

Input

Context file: acceptance-context.json

{
  "issue_number": 166,
  "feature_summary": "Feature summary",
  "acceptance_criteria": ["Criterion 1", "Criterion 2"],
  "test_scenarios": ["Scenario 1", "Scenario 2"]
}

Output

Result file: acceptance-result.json

{
  "status": "passed",
  "test_cases": [
    { "scenario": "Scenario 1", "result": "passed" },
    { "scenario": "Scenario 2", "result": "passed" }
  ],
  "acceptance_criteria_status": [
    { "criterion": "Criterion 1", "verified": true },
    { "criterion": "Criterion 2", "verified": true }
  ]
}

Completion Criteria

  • All test scenarios pass
  • All acceptance criteria verified
  • Evidence collected

refactoring-agent

Overview

A specialist agent that improves code quality, applies design patterns based on SOLID principles, and resolves technical debt.

Input

Context file: refactor-context.json

{
  "issue_number": 166,
  "refactor_targets": ["src/lib/xxx.ts"],
  "quality_metrics": {
    "before_coverage": 75.0
  },
  "improvement_goals": [
    "Increase coverage to 80%+",
    "Remove duplicate code"
  ]
}

Output

Result file: refactor-result.json

{
  "status": "success",
  "quality_metrics": {
    "before_coverage": 75.0,
    "after_coverage": 82.0
  },
  "refactorings_applied": [
    "Split long functions",
    "Removed duplicate code"
  ],
  "files_changed": ["src/lib/xxx.ts"]
}

Completion Criteria

  • All tests pass
  • Quality metrics improved
  • Zero static analysis errors

How to Call Agents

Calling from PM Auto-Dev

Use tdd-impl-agent to implement Issue #166 with TDD approach.

Context file: dev-reports/issue/166/pm-auto-dev/iteration-1/tdd-context.json
Output file: dev-reports/issue/166/pm-auto-dev/iteration-1/tdd-result.json

Calling from PM Bug Fix

Use investigation-agent to investigate the bug.

Context file: dev-reports/bug-fix/20251205_120000/investigation-context.json
Output file: dev-reports/bug-fix/20251205_120000/investigation-result.json

File Structure

.claude/
├── agents/
│   ├── tdd-impl-agent.md
│   ├── progress-report-agent.md
│   ├── investigation-agent.md
│   ├── acceptance-test-agent.md
│   └── refactoring-agent.md
└── prompts/
    ├── tdd-impl-core.md
    ├── progress-report-core.md
    ├── refactoring-core.md
    └── acceptance-test-core.md

dev-reports/
├── issue/{issue_number}/
│   └── pm-auto-dev/
│       └── iteration-{N}/
│           ├── tdd-context.json
│           ├── tdd-result.json
│           ├── acceptance-context.json
│           ├── acceptance-result.json
│           ├── refactor-context.json
│           ├── refactor-result.json
│           ├── progress-context.json
│           └── progress-report.md
└── bug-fix/{bug_id}/
    ├── investigation-context.json
    ├── investigation-result.json
    └── ...

Related Documentation