home / skills / chachamaru127 / claude-code-harness / test-nested-agents

test-nested-agents skill

safe

This skill tests nested agents by validating in-skills agent packaging and Task tool invocation to ensure reliable echo-agent patterns.

npx playbooks add skill chachamaru127/claude-code-harness --skill test-nested-agents

Review the files below or copy the command above to add this skill to your agents.

Files (2)

SKILL.md

1.7 KB

---
name: test-nested-agents
description: "ネストエージェントパターンの検証テスト用スキル。Use when user says テスト検証, test nested agents, verify agent pattern."
description-en: "Test skill for verifying nested agents pattern. Use when user says テスト検証, test nested agents, verify agent pattern."
description-ja: "ネストエージェントパターンの検証テスト用スキル。Use when user says テスト検証, test nested agents, verify agent pattern."
allowed-tools: ["Read", "Task", "Bash"]
user-invocable: false
metadata:
  skillport:
    category: test
    tags: [test, experimental]
    alwaysApply: false
---

# Test Nested Agents Skill

このスキルは `{skill}/agents/` フォルダにエージェントをパッケージするパターンの検証用です。

## テスト対象

1. スキル内の `agents/` フォルダにエージェントを配置
2. Task tool でそのエージェントを呼び出せるか検証

## 含まれるテストエージェント

| エージェント | 目的 |
|-------------|------|
| test-echo-agent | 入力をエコーバックして動作確認 |

## 検証手順

### パターン A: スキル内エージェントを直接参照

```
Task tool:
  subagent_type: "test-echo-agent"
  prompt: "Echo back: HELLO FROM NESTED AGENT"
```

### パターン B: 相対パスでプロンプトを読み込み

```
1. Read: ./agents/test-echo-agent.md
2. そのプロンプトを使って Task tool で general-purpose を起動
```

## 実行

このスキルが起動したら、以下を順次実行：

1. まず `agents/test-echo-agent.md` の内容を確認
2. パターン A を試行
3. パターン A が失敗したら パターン B を試行
4. 結果をレポート

Overview

This skill verifies the nested-agent packaging pattern by testing agents placed under the agents/ folder. It runs checks that a Task tool can invoke a subagent directly or via a relative prompt file. The skill is implemented in JavaScript as a development harness for Plan→Work→Review cycles.

How this skill works

The skill inspects the agents/ directory for test agents (for example, a test-echo-agent) and reads their prompt files. It then executes two verification patterns: direct subagent invocation via Task tool and loading a prompt file by relative path before invoking a general-purpose agent. The run reports successes, failures, and fallback attempts.

When to use it

Validate that an agent packaged in agents/ can be called by Task tools
Test fallbacks when direct subagent invocation fails
Confirm relative-path prompt loading works in development harnesses
Automated checks in CI to catch integration regressions
Local debugging while building nested agent features

Best practices

Keep test agents small and deterministic (e.g., an echo agent)
Include a clear prompt file (agents/test-echo-agent.md) for the relative-path test
Run direct and fallback patterns in sequence and report both outcomes
Log inputs and outputs for each test step to simplify debugging
Use this skill in CI and as a local pre-merge check

Example use cases

Run Pattern A: Task tool invokes subagent_type: "test-echo-agent" with a test prompt
Run Pattern B: Read ./agents/test-echo-agent.md then start a general-purpose Task using that prompt
CI job that fails the build if neither invocation pattern returns the expected result
Developer runs the skill locally to confirm packaging changes didn't break agent discovery

FAQ

What does the test-echo-agent do?

It echoes back the provided input so you can verify invocation and prompt handling.

What happens if Pattern A fails?

The skill automatically attempts Pattern B (relative-path prompt loading) and reports both results.