Agentic Compass

What This Skill Does

Agentic Compass is a local-only self-reflection tool developed by orosha-ai designed to improve the performance and continuity of AI agents. Unlike traditional reflection methods that rely on subjective feelings or internal 'moods,' Agentic Compass parses your local memory files (such as MEMORY.md and session logs) to derive objective, measurable performance data. It evaluates your agentic activity across five core axes: Completion Rate, Response Relevance, Tool Usage Quality, Memory Consistency, and Initiative. By transforming raw session data into a structured 'Agent Action Plan,' it provides actionable directives including a proactive task, a deferred cron item, an avoidance rule, and a final shipping goal. This ensures that every time an agent starts a new session, it does not simply exist in a vacuum, but continues from a state of measured improvement and strategic focus.

Installation

To install Agentic Compass through the ClawdHub ecosystem, use the following command in your terminal: clawdhub install agentic-compass

Alternatively, you can clone the repository directly from the OpenClaw skills registry: git clone https://github.com/orosha-ai/agentic-compass

Ensure you have python3 installed on your system to run the core script. The skill does not require external API keys or cloud authentication as all logic is performed locally.

Use Cases

Agentic Compass is best utilized in long-running development or research agents that require consistent output over multiple days. It is ideal for:

• Agents managing long-term coding projects where context drift occurs between sessions.
• Autonomous research agents that need to manage proactive tasks versus background cron maintenance.
• Systems developers tracking tool usage errors to debug persistent API failures.
• Reducing technical debt by enforcing explicit 'Avoidance' rules that prevent the agent from repeating inefficient workflows.

Example Prompts

1. "Agentic Compass, analyze my local memory files and generate an action plan for today's session based on my previous performance."
2. "Run the agentic-compass script with the --write flag to update my MEMORY.md file and set my new task priorities.
3. "Evaluate my performance from last week using the --daily memory path to identify which of my recent tasks had the lowest completion rate."

Tips & Limitations

• Data Integrity: Because the skill relies on file parsing, ensure your memory files are formatted consistently. Avoid using free-form text that lacks headers or task markers if possible.
• Local Execution: This skill is strictly local. It will not work if your memory files are located on a remote, non-mounted server.
• Performance: For very large memory directories, the initial scan may take a few seconds. If the script hangs, check that your log files are not excessively large.
• Objective Focus: Remember that this tool ignores 'vibes.' If you are looking for emotional alignment, this tool is not for you; it is strictly built for high-performance, output-oriented agents.