Memory Health Score

Provides a comprehensive score (0-100) evaluating the Agent's memory system health across completeness, freshness, structure, density, and consistency.

This skill appears coherent and limited to reading/writing files in the agent workspace and producing a JSON health report — it does not make network calls or ask for credentials. Before installing or enabling autonomous runs: 1) review check-health.mjs to confirm its filesystem targets are acceptable (it will read MEMORY.md, memory/*.md, .issues, HEARTBEAT.md and write memory/health-score.json); 2) be aware SKILL.md mentions automatic repair and dashboard UI, but those are not implemented in the provided script — do not assume the skill will auto-modify your memory files; 3) if you plan to run it on a sensitive workspace, run it first in an isolated session or inspect the repo for any additional repair scripts it references; 4) consider whether you want the agent to invoke this autonomously (normal default) or only run it manually via the cron command.

Type: OpenClaw Skill Name: memory-health-score Version: 1.0.0 The skill is classified as suspicious due to potential prompt injection vectors in the markdown files. Specifically, SKILL.md contains instructions for the AI agent to 'automatically trigger' memory compression, index rebuilding, and task cleaning if the health score is low. While these actions are related to the skill's stated purpose of memory health, they instruct the agent to perform further actions without explicit user confirmation, which is a risky capability. Additionally, agent-amnesia-guide.md, a documentation file, includes shell commands (e.g., `git`, `cat | grep`, `node` calls to other local scripts) as 'Acute Treatment' suggestions. An AI agent could potentially interpret and execute these commands if prompted to 'fix' its memory, posing a risk of unintended execution, even though the commands themselves are related to memory management and not overtly malicious.