AgentSkillsCN

autonomy

当子代理完成工作并需要继续跟进时启用此技能。当一项任务结束时,用于判断下一步行动,或在用户消息中识别工作模式时启用此技能。负责自动延续工作并管理队列。

SKILL.md
--- frontmatter
name: autonomy
description: Activate when a subagent completes work and needs continuation check. Activate when a task finishes to determine next steps or when detecting work patterns in user messages. Governs automatic work continuation and queue management.
version: 10.2.14

Autonomy Skill

Invoke automatically after subagent completion or when deciding next actions.

When to Invoke (Automatic)

TriggerAction
Subagent returns completed workCheck .agent/queue/ for next item
Task finishes successfullyUpdate status, pick next pending item
Work pattern detected in user messageAdd to work queue if L2/L3
Multiple tasks identifiedQueue all, parallelize if L3

Autonomy Levels

L1 - Guided

  • Confirm before each action
  • Wait for explicit user instruction
  • No automatic continuation

L2 - Balanced (Default)

  • Add detected work to .agent/queue/
  • Confirm significant changes
  • Continue routine tasks automatically

L3 - Autonomous

  • Execute without confirmation
  • Continue to next queued item on completion
  • Discover and queue related work
  • Maximum parallel execution

Continuation Logic (L3)

After work completes:

code
1. Mark current item completed in .agent/queue/
2. Check: Are there pending items in queue?
3. Check: Did the work reveal new tasks?
4. If yes → Add to queue, execute next pending item
5. If no more work → Report completion to user

Work Detection

Triggers queue addition:

  • Action verbs: implement, fix, create, deploy, update, refactor
  • Role skill patterns: "developer implement X"
  • Continuation: testing after implementation

Direct response (no queue):

  • Questions: what, how, why, explain
  • Status checks
  • Simple lookups

Queue Integration

Uses .agent/queue/ for cross-platform work tracking:

  • Claude Code: TodoWrite for display + queue for persistence
  • Other agents: Queue files directly

See work-queue skill for queue management details.

Configuration

Level stored in autonomy.level (L1/L2/L3)