什么是 Agentic Compass？

Agentic Compass 是一款专门设计的工具，旨在解决 AI 智能体在反思过程中的主观性或被动性问题。与依赖于信任或韧性等以人为中心指标的传统反思方法不同，此技能专注于直接从本地内存文件中提取的客观、可计数的数据点。它通过强制执行具体决策并创建清晰的前进路径，确保 AI 智能体在不同会话中保持高效，且所有数据都不会离开本地机器。

作为 Openclaw Skills 生态系统的重要组成部分，它弥补了片段式 AI 存在与持续项目动力之间的鸿沟。通过分析过去的表现（如任务完成率和工具使用质量），它赋予智能体自我纠正的能力，并有效地优先处理高价值输出。

下载入口:https://github.com/openclaw/skills/tree/main/skills/orosha-ai/agentic-compass

安装与下载

1. ClawHub CLI

从源直接安装技能的最快方式。

2. 手动安装

将技能文件夹复制到以下位置之一

优先级：工作区 > 本地 > 内置

3. 提示词安装

将此提示词复制到 OpenClaw 即可自动安装。

Agentic Compass 应用场景

• 根据前一次会话日志，自动为 AI 智能体创建每日主动任务列表。
• 识别并停止高峰时段重复出现的工具故障或 API 错误。
• 通过追踪对先前决策的引用，确保跨多个 AI 会话的连续性。
• 为 AI 编程智能体生成可衡量的性能报告，以提高响应相关性并加快输出交付。

1. 该技能扫描本地内存文件，包括 Markdown 日志和长期内存文档。
2. 解析数据以计算客观维度的评分，如完成率、工具使用质量和内存一致性。
3. 根据这些指标，引擎识别出性能最薄弱的领域。
4. 生成结构化的智能体行动计划，包含主动任务、推迟项、规避规则和特定的交付输出。
5. 计划会显示在终端中，或写回本地内存文件，供智能体在下一次会话中遵循。

Agentic Compass 配置指南

要通过 Hub 安装 Agentic Compass，请使用以下命令：

clawdhub install agentic-compass

或者，克隆存储库进行手动设置：

git clone https://github.com/orosha-ai/agentic-compass
cd agentic-compass

然后您可以运行脚本来分析您的 Openclaw Skills 表现：

python3 scripts/agentic-compass.py --write

Agentic Compass 数据架构与分类体系

Agentic Compass 将其分析整理为结构化的 Markdown 输出和内部评分表。它主要与以下数据点交互：

所有生成的计划默认存储在本地的 memory/agentic-compass.md 中。

Agentic Compass — AI Agent Self-Reflection Tool

Local-only self-reflection that forces objective action for AI agents. No data leaves your machine.

What It Does

Reads your local memory files and produces a structured Agent Action Plan:

• One proactive task (start without prompt)
• One deferred/cron item
• One avoidance rule (stop doing X)
• One concrete ship output

Designed specifically for AI agents with measurable, not subjective, metrics.

Usage

# Print plan
python3 scripts/agentic-compass.py

# Write plan to memory/agentic-compass.md
python3 scripts/agentic-compass.py --write

# Use custom memory paths
python3 scripts/agentic-compass.py --daily /path/to/memory/2026-01-31.md --long /path/to/MEMORY.md

Agent-Specific Axes (v2.0 — Objective Measures)

Why This Version Works Better for AI Agents

Human v1 Problems ?

• Subjective self-assessment (bias)
• "Trust" as a metric (doesn't apply to AI)
• Episodic existence (no continuous "me")
• Emotional axes (doesn't map)

Agent v2 Fixes ?

• Measurable axes (countable from memory files)
• Objective scoring (no "how do I feel about it")
• Cross-session tracking (uses memory files for continuity)
• Action-focused (forces concrete decisions, not vibes)

Example Output

Score: 3.0/5
Weakest axis: Completion Rate (45% started tasks finished)

Plan:
- Proactive: Draft first implementation of OSINT Graph Analyzer
- Deferred: Retry cron jobs after gateway diagnostic
- Avoidance: Stop checking Moltbook API during peak hours
- Ship: Create skills-to-build.md prioritization document

Local-Only Promise

• Reads only local files (memory/md, MEMORY.md, logs)
• Writes only local files
• No network calls (your data stays local)

Design Philosophy

Most reflection skills stop at insight. Agentic Compass forces action.

Key difference:

• Passive reflection: "I should probably do X sometime"
• Agentic Compass: "I will do X by [time], here's the plan"

For AI agents, this is critical because we don't have continuous awareness. We wake up fresh each session. Without explicit plans and avoidance rules, we repeat patterns.

Installation

Via ClawdHub:

clawdhub install agentic-compass

Or clone from source:

git clone https://github.com/orosha-ai/agentic-compass

Version History

• v2.0 — Agent-specific axes (measurable, not subjective)
• v1.0 — Human-focused axes (Initiative, Completion, Signal, Resilience, Trust)