AgentSkillsCN

parallel-agents

针对 3 个以上彼此独立且无共享状态的故障或任务，派遣并行子代理。

SKILL.md
--- frontmatter
name: parallel-agents
description: Dispatch parallel subagents for 3+ independent failures or tasks with no shared state

Parallel Agents

Dispatch one agent per independent problem domain. Let them work concurrently.

When to Use

  • •3+ failures in different test files/subsystems
  • •Each problem understood without context from others
  • •No shared state between investigations

When NOT to Use

  • •Related failures (fix one → might fix others)
  • •Need full system context
  • •Agents would interfere (same files, shared resources)
  • •Exploratory debugging (don't know what's broken yet)

Process

  1. •Group by domain: Identify independent problem areas
  2. •Dispatch focused agents: One Task per domain
  3. •Review + integrate: Read summaries, verify no conflicts, run full suite

Agent Prompt Structure

Good prompts are focused, self-contained, specific about output:

code
Fix 3 failing tests in src/foo/bar.test.ts:

1. "test name A" - expected X got Y
2. "test name B" - timeout waiting for Z
3. "test name C" - wrong count

Root cause likely: timing/race conditions

Task:
1. Read test file, understand what each verifies
2. Identify root cause - timing issues or bugs?
3. Fix with event-based waiting, not arbitrary timeouts

Do NOT just increase timeouts.

Return: Summary of findings + fixes.

Prompt Anti-patterns

  • •Too broad ("Fix all tests") → agent gets lost
  • •No context ("Fix race condition") → where?
  • •No constraints → agent refactors everything
  • •Vague output ("Fix it") → don't know what changed

Verification

After agents return:

  1. •Review each summary
  2. •Check for conflicts (same code edited?)
  3. •Run full test suite
  4. •Spot check (agents can make systematic errors)