技能 dispatching-parallel-agents
📦

dispatching-parallel-agents

安全

Отправка параллельных агентов для более быстрой отладки

也可从以下获取: Doyajin174,DYAI2025,obra,Cycleaddict,ZhanlinCui,Cygnusfear,davila7,CodingCossack

Когда сталкиваетесь с несколькими независимыми сбоями тестов, их последовательное исследование теряет время. Этот навык научит вас отправлять одного AI-агента на каждую проблемную область, решая несколько ошибок параллельно.

支持: Claude Codex Code(CC)
📊 71 充足
1

下载技能 ZIP

2

在 Claude 中上传

前往 设置 → 功能 → 技能 → 上传技能

3

开启并开始使用

测试它

正在使用“dispatching-parallel-agents”。 We have 6 test failures across 3 files after a major refactor

预期结果:

I'll identify independent problem domains:

1. agent-tool-abort.test.ts: 3 failures (timing issues)
2. batch-completion-behavior.test.ts: 2 failures (tools not executing)
3. tool-approval-race-conditions.test.ts: 1 failure (execution count = 0)

These are independent domains. Dispatching 3 agents in parallel to fix each file.

正在使用“dispatching-parallel-agents”。 Integrate the results from our 3 parallel agents

预期结果:

All 3 agents completed successfully:
- Agent 1: Replaced timeouts with event-based waiting
- Agent 2: Fixed event structure bug
- Agent 3: Added wait for async execution

No conflicts detected. Full test suite is green. Time saved: 3 problems solved in parallel vs sequentially.

安全审计

安全
v1 • 2/24/2026

Security audit complete. All 10 static findings are false positives. The skill is documentation about parallel agent dispatching methodology. Backtick patterns detected are markdown code fences, not shell execution. No cryptographic algorithms, network requests, or system reconnaissance present. Safe for marketplace publication.

1
已扫描文件
183
分析行数
0
发现项
1
审计总数
未发现安全问题
审计者: claude

质量评分

38
架构
100
可维护性
87
内容
22
社区
100
安全
100
规范符合性

你能构建什么

Сбои в нескольких тестовых файлах

Когда 3 или более тестовых файлов fail с разными root causes после рефакторинга, отправьте отдельных агентов для одновременного исправления каждого файла.

Ошибки независимых подсистем

Когда разные подсистемы ломаются независимо (например, auth, caching, API), назначьте каждую отдельному агенту для параллельного исследования.

Сбои в CI/CD pipeline

Когда несколько этапов pipeline fail по не связанным причинам, распараллельте исследование между агентами, чтобы быстрее достичь зелёного статуса.

试试这些提示

Базовая параллельная отправка 
I have multiple independent test failures. Please analyze and dispatch appropriate agents to fix each one:

1. [Failure 1 - describe file and error]
2. [Failure 2 - describe file and error]
3. [Failure 3 - describe file and error]

These are independent problems. Return summaries of what each agent finds and fixes.
Focused Agent Task 
Fix the failing tests in [test-file-name.ts]:

1. [Test name 1] - [expected behavior vs actual]
2. [Test name 2] - [expected behavior vs actual]

This is a [timing issue / bug / race condition]. Your task:
1. Read and understand the test file
2. Identify root cause
3. Fix the issue
4. Return summary of what you found and changed.
Интеграция результатов 
Three agents have completed fixes for:
- Agent 1: [file A] - [summary]
- Agent 2: [file B] - [summary]
- Agent 3: [file C] - [summary]

Please:
1. Check if any fixes conflict
2. Run the full test suite
3. Verify all changes work together
4. Report final status
Запрос на параллельное исследование 
I need to investigate [number] independent issues. For each one, create a focused agent task:

Issue 1: [problem description]
Issue 2: [problem description]
Issue 3: [problem description]

Dispatch all agents in parallel. Wait for all to complete, then help me integrate the results.

最佳实践

  • Always verify problems are truly independent before dispatching parallel agents
  • Give each agent specific scope, clear goals, and constraints on what to change
  • Require agents to return structured summaries for easier integration
  • Run full test suite after all agents complete to verify no conflicts

避免

  • Dispatching parallel agents for related failures that might share fixes
  • Creating agent tasks too broad like 'fix all the tests' without specific scope
  • Not providing enough context for agents to understand the problem domain
  • Skipping the integration step - always verify fixes work together

常见问题

Как мне узнать, достаточно ли независимы проблемы для параллельных агентов?
Problems are independent if fixing one does not affect the others. If fixing a bug in module A would also fix a bug in module B, they are related. Also check for shared state - if agents would edit the same files, they are not independent.
Какое минимальное количество сбоев для использования этого навыка?
This skill becomes valuable with 2+ independent failures. With 3+ failures, the time savings are significant. For single failures, use standard sequential debugging.
Могут ли агенты вмешиваться друг в друга при параллельной работе?
Yes, if they share state like editing the same files, using the same database, or modifying shared configuration. Always verify independence before dispatching.
Как правильно структурировать задачу агента для параллельной отправки?
A good agent task needs: (1) Specific scope - one file or subsystem, (2) Clear goal - what tests should pass, (3) Constraints - what not to change, (4) Expected output - what to return.
Что делать, если агенты возвращают конфликтующие исправления?
Review both changes to understand the conflict. You may need to pick one fix, modify both to work together, or investigate further. Run the full test suite to verify the resolution.
Работает ли этот навык с Claude Code и Codex?
Yes, this methodology works with any AI agent system including Claude Code, Codex, and other AI coding assistants. The pattern is about dispatching multiple agent tasks concurrently.

开发者详情

作者

sickn33

许可证

MIT

仓库

https://github.com/sickn33/antigravity-awesome-skills/tree/main/skills/dispatching-parallel-agents

引用

main

文件结构

📄 SKILL.md