返回 Skill 列表
extension
分类: 开发与工程无需 API Key

agent-patterns

当用户询问“SPAWN REQUEST格式”、“代理报告”、“代理协调”、“并行代理”、“报告格式”、“代理通信”,或需要了解代理如何在冲刺系统内协调时,应使用此技能。在检测到适当上下文时使用。根据技能目的通过相关短语触发。

person作者: jakexiaohubgithub

Agent Patterns

Coordinate multi-agent workflows using structured communication protocols. Covers spawn requests, status reports, and inter-agent messaging within the sprint system.

Spawn Request Format

When launching a parallel agent, use this structure:

## SPAWN REQUEST
- **Agent ID**: <unique-id>
- **Role**: <role-name>
- **Task**: <one-line description>
- **Context**: <relevant files or state>
- **Expected Output**: <deliverable format>
- **Timeout**: <max duration>

Agent Report Format

Each agent reports back using:

## AGENT REPORT
- **Agent ID**: <id>
- **Status**: completed | failed | blocked
- **Summary**: <one-line result>
- **Output**: <deliverable or file path>
- **Issues**: <blockers or warnings, if any>
- **Duration**: <time taken>

Coordination Workflow

  1. Define tasks — break work into independent units suitable for parallel execution
  2. Spawn agents — send SPAWN REQUEST for each task with clear context
  3. Monitor status — track agent reports as they complete
  4. Handle failures — retry failed agents or reassign blocked tasks
  5. Merge results — combine agent outputs into final deliverable, resolving conflicts

Communication Rules

  • Agents communicate through structured reports, not free-form text
  • Each message includes agent ID for traceability
  • Blocked agents must report immediately rather than waiting silently
  • The coordinator agent reviews all reports before merging

Error Handling

See references/errors.md for common failure modes and recovery steps.

Examples

See references/examples.md for complete spawn-and-report workflows.