agentic-compass
由 orosha-ai 提供
AI & LLMs简介
Local-only self-reflection that forces action for AI agents.
描述
该技能的功能
Reads local memory files to score an AI agent across five objective axes and outputs a four-part action plan: one proactive task, one deferred item, one avoidance rule, and one concrete deliverable. All processing is local, no network calls are made.
Uses countable signals from memory files (done markers, error logs, correction counts) rather than subjective self-rating, so scores reflect actual behavior rather than agent bias.
何时使用
- Auditing task completion rate after a long session
- Catching repeated tool failures logged in memory files
- Generating a proactive first task to run without a user prompt
- Writing an avoidance rule to stop a recurring bad pattern
- Scheduling a deferred job based on current memory state
示例工作流
User asks: What should I focus on this session?
- 1
Run agentic-compass.py to read daily and long-term memory files
- 2
Score each axis by parsing markers and logs in those files
- 3
Identify the weakest axis from the scored results
- 4
Produce a four-part plan: proactive task, deferred item, avoidance rule, ship output
- 5
Optionally write the plan to memory/agentic-compass.md with --write flag
Score: 3.0/5. Weakest axis: Completion Rate. Plan includes drafting a specific implementation, retrying deferred cron jobs, avoiding a named API during peak hours, and creating a prioritization document.
安装
clawhub install orosha-ai/agentic-compass运行 CLI 命令,或把这段提示词粘贴到 Claude Code、Cursor、Codex 或你在用的任意编程智能体里。