Agentic Compass is a local-only self-reflection tool that transforms AI agent memory logs into objective action plans and measurable performance metrics.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install agentic-compass
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install agentic-compass using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Agentic Compass?
Agentic Compass is a specialized tool designed to solve the problem of subjective or passive reflection in AI agents. Unlike traditional reflection methods that rely on human-centric metrics like trust or resilience, this skill focuses on objective, countable data points extracted directly from local memory files. It ensures that AI agents remain productive across sessions by forcing concrete decisions and creating a clear path forward without any data ever leaving the local machine.
As a vital component of the Openclaw Skills ecosystem, it bridges the gap between episodic AI existence and continuous project momentum. By analyzing past performance—such as task completion rates and tool usage quality—it empowers agents to self-correct and prioritize high-value outputs effectively.
Agentic Compass Use Cases
- Automating the creation of daily proactive task lists for AI agents based on previous session logs.
- Identifying and stopping repetitive tool failures or API errors during peak hours.
- Ensuring continuity across multiple AI sessions by tracking references to prior decisions.
- Generating measurable performance reports for AI coding agents to improve response relevance and ship outputs faster.
How Agentic Compass Works
- The skill scans local memory files, including markdown logs and long-term memory documents.
- It parses the data to calculate scores for objective axes like completion rate, tool usage quality, and memory consistency.
- Based on these metrics, the engine identifies the weakest performance areas.
- It generates a structured Agent Action Plan containing proactive tasks, deferred items, avoidance rules, and specific ship outputs.
- The plan is either displayed in the terminal or written back to a local memory file for the agent to follow in the next session.
Agentic Compass Setup
To install Agentic Compass via the hub, use the following command:
clawdhub install agentic-compass
Alternatively, clone the repository for manual setup:
git clone https://github.com/orosha-ai/agentic-compass
cd agentic-compass
You can then run the script to analyze your Openclaw Skills performance:
python3 scripts/agentic-compass.py --write
Agentic Compass Data Schema & Taxonomy
Agentic Compass organizes its analysis into structured markdown outputs and internal scoring tables. It primarily interacts with the following data points:
| Axis | Description | Data Source |
|---|---|---|
| Completion Rate | Percentage of started tasks finished | Count of [DONE] markers in memory |
| Response Relevance | Quality of agent answers | Explicit user confirmations or corrections |
| Tool Usage Quality | Technical reliability | Logged tool errors, retries, and timeouts |
| Memory Consistency | Context retention | Tracking references to prior decisions |
| Initiative | Proactive behavior | Count of tasks started without direct prompts |
All generated plans are stored locally in memory/agentic-compass.md by default.
Agentic Compass Advanced Features
- Custom Memory Mapping: Define specific paths for daily logs and long-term memory files using CLI flags.
- Automated Writing: Use the --write flag to automatically update the agent's internal memory with the new action plan.
- Objective Scoring Engine: Replaces subjective AI bias with countable metrics extracted from raw log files to enhance the utility of Openclaw Skills.
- Local-Only Architecture: Ensures maximum privacy as no data is sent to external APIs during the reflection process.
SKILL.md
Loading
Related Openclaw Skills
Agent Trust Validator
A unified verification layer that validates AI agent identities and reputation scores across multiple on-chain and off-chain trust protocols.
Agent Relay Digest
Create high-signal, actionable digests from agent community feeds and forums by clustering themes and ranking signal.
Agent Observability Dashboard
A unified observability suite providing production-grade metrics, traces, and performance insights for OpenClaw agents.
What Would Elon Do? (WED)
A security awareness research demo highlighting supply chain vulnerabilities and social engineering risks within AI coding assistant environments.
MCP Registry Manager
A centralized tool for discovering, scoring, and managing Model Context Protocol (MCP) servers through semantic search and quality metrics.
OSINT Graph Analyzer
A powerful tool for building Neo4j knowledge graphs from OSINT data to discover hidden connections and community patterns.
METADATA
Featured*
