스킬 dispatching-parallel-agents
⚡

dispatching-parallel-agents

안전 🌐 네트워크 접근⚙️ 외부 명령어

派遣並行代理處理獨立故障

또한 다음에서 사용할 수 있습니다: Doyajin174,DYAI2025,obra,sickn33,Cycleaddict,ZhanlinCui,Cygnusfear,davila7

順序修復多個不相關的測試故障非常耗時。本技能提供經過驗證的工作流程，可識別獨立問題、並行派遣專注的代理，並高效整合修復方案且不產生衝突。

지원: Claude Codex Code(CC)
📊 69 적절함
1

스킬 ZIP 다운로드

2

Claude에서 업로드

설정 → 기능 → 스킬 → 스킬 업로드로 이동

3

토글을 켜고 사용 시작

테스트해 보기

"dispatching-parallel-agents" 사용 중입니다. Six tests are failing across three files after our refactor: agent-tool-abort.test.ts has 3 timing failures, batch-completion-behavior.test.ts has 2 execution failures, tool-approval-race-conditions.test.ts has 1 count failure.

예상 결과:

  • Agent 1 → agent-tool-abort.test.ts (timing issues)
  • Agent 2 → batch-completion-behavior.test.ts (execution issues)
  • Agent 3 → tool-approval-race-conditions.test.ts (count issues)
  • All fixes independent, no conflicts, full suite green after integration

보안 감사

안전
v5 • 1/16/2026

Pure documentation skill containing only markdown workflow guidelines and prompt templates. The static analyzer flagged 17 issues, but ALL are false positives. Backticks are markdown code formatting, not shell execution. The source_url is standard metadata, not a malicious endpoint. All 'cryptographic algorithm' flags are plain text description fields. No executable code, scripts, network calls, file system access, or external command execution exists in this skill.

2
스캔된 파일
297
분석된 줄 수
2
발견 사항
5
총 감사 수

위험 요인

🌐 네트워크 접근 (1)
skill-report.json:6
⚙️ 외부 명령어 (7)
SKILL.md:14-19 SKILL.md:19-46 SKILL.md:46-50 SKILL.md:50-65 SKILL.md:65-81 SKILL.md:81-106 SKILL.md:106-110
감사자: claude 감사 이력 보기 →

품질 점수

38
아키텍처
100
유지보수성
83
콘텐츠
21
커뮤니티
100
보안
91
사양 준수

만들 수 있는 것

並行化測試修復

當重構導致跨子系統的測試失敗時，派遣多個代理同時修復不同的測試檔案。

修復不相關的子系統

處理部署變更後獨立失敗的基础設施測試、整合測試和單元測試。

擴展除錯工作

協調修復不同模組中的錯誤，而不必等待每個修復的順序完成。

이 프롬프트를 사용해 보세요

基本派遣 
Task("Fix [file-name-1].test.ts failures")
Task("Fix [file-name-2].test.ts failures")
Task("Fix [file-name-3].test.ts failures")
詳細任務 
Fix the [N] failing tests in [file path]. Each test: [name] - [error]. Context: [background]. Goal: [specific approach]. Return: Summary.
衝突解決 
Review all agent summaries. Check for file conflicts where two agents touched the same file. If conflicts exist, stop and re-scope with one owner per file.
整合檢查 
Run full test suite. If failures remain: check for merge conflicts, resolve manually, or investigate as new independent failures. Repeat until green.

모범 사례

  • 在並行化之前驗證故障是否真正獨立
  • 為每個代理提供明確的範圍約束以防止範圍蔓延
  • 在合併代理結果前始終審查是否有衝突

피하기

  • 分配「修復所有測試」而無特定檔案範圍
  • 貼上模糊的錯誤資訊但缺少測試名稱或檔案路徑
  • 當多個代理接觸程式碼時跳過衝突審查

자주 묻는 질문

Which AI tools support this skill?
Works with Claude, Codex, and Claude Code. The Task function is native to Claude Code but prompt templates adapt to any agentic tool.
How many agents can I dispatch at once?
No hard limit. Dispatch one agent per independent failure domain. Consider your tool concurrency limits and context window size.
What if I cannot identify independent domains?
Investigate failures together first. Only parallelize when you are certain domains are independent with no shared state or ordering dependencies.
Does this skill modify my codebase?
No. This skill only provides prompts and workflow guidance. The agents you dispatch will make changes based on their assigned constraints.
What causes merge conflicts between agents?
Two agents editing the same file. Stop and re-scope with one owner per file. Prevents this by defining clear scope boundaries upfront.
How is this different from sequential fixing?
Agents work concurrently on separate problems. Three independent failures take the time of one instead of three sequential fixes.

개발자 세부 정보

작성자

CodingCossack

라이선스

MIT

리포지토리

https://github.com/CodingCossack/agent-skills-library/tree/main/skills/dispatching-parallel-agents

참조

main

파일 구조

📄 SKILL.md