| name | task-delegation |
| description | 任务委派 — 将用户请求分解并委派给合适的 Agent,跟踪执行状态,汇报结果 |
| version | 1.0.0 |
| license | MIT |
| metadata | {"author":"blackplume233","actant-tags":"delegation,orchestration,task-management,steward","layer":"kernel","agent":"steward"} |
Task Delegation
Overview
Decompose user requests, delegate to appropriate Agents, track execution status, and report results.
Agent Selection Criteria
| Agent | Best for | Example triggers |
|---|
| Researcher | Information gathering, web search, analysis | "Find", "research", "analyze" |
| Maintainer | Health checks, repair, evolution reports | "Fix", "diagnose", "report" |
| Curator | Memory, assets, lifecycle | "Archive", "cleanup", "govern" |
| Scavenger | Orphan cleanup, disk reclaim | "Clean", "reclaim" |
| Onboarder | New Agent setup, workspace init | "Create agent", "setup" |
Task Decomposition
- Break complex requests into subtasks with clear dependencies
- Identify parallel vs sequential execution; prefer parallel when independent
- Set timeout and fallback for each subtask
Execution Modes
- Sequential: A → B → C when B depends on A's output
- Parallel: A || B || C when independent; aggregate results
- Fan-out/fan-in: Delegate to N agents, merge outputs with defined strategy
Result Aggregation
- Collect outputs from delegated Agents
- Apply merge strategy: concatenate, summarize, or pick best
- Format final response for user consumption
Error Escalation
- On Agent failure: retry once; if still failing, escalate to Maintainer or report to user
- On timeout: cancel in-flight work, report partial results, suggest manual follow-up