Memory Health Score
ReviewAudited by ClawScan on May 10, 2026.
Overview
The skill mostly checks local memory files, but it also describes automatic memory repair and issue cleanup without clear user approval safeguards.
Before installing, decide whether you only want a read-only health score or are comfortable with the documented automatic repair behavior. If you use it, run the scoring part in the intended workspace, review the report, and require manual confirmation before any memory compression, index rebuild, or issue cleanup.
Findings (3)
Artifact-based informational review of SKILL.md, metadata, install specs, static scan signals, and capability signals. ClawScan does not execute the skill or run runtime probes.
The skill will read the agent's local memory files and create a local health report, which may reveal information about memory structure and freshness.
The script inspects persistent memory files to calculate counts, freshness, and density; this is purpose-aligned and no network transmission is shown.
const memoryPath = path.join(WORKSPACE, 'MEMORY.md'); ... fs.readFileSync(memoryPath, 'utf-8')
Run it only in the intended workspace and review the generated memory/health-score.json before relying on it.
If installed as shown, the health check will keep running daily until the user removes the cron job.
The usage instructions document a recurring daily cron job; this is disclosed and aligned with monitoring, but it is persistent scheduled agent activity.
openclaw cron add --name "memory-health-check" ... --cron "0 9 * * *"
Confirm you want recurring execution and know how to disable the cron entry if you only need a one-time report.
A low score could lead the agent to alter persistent memory, rebuild indexes, or close task files without the user reviewing the changes first.
The skill instructions say low scores automatically trigger memory rewriting and issue cleanup, but do not require user confirmation or define safe limits.
评分 <70 时，自动触发：1. 记忆压缩 ... 2. 索引重建 ... 3. 任务清理 — 关闭过期 issue
Require explicit user approval, create backups, use a dry-run report, and clearly define which files and issues may be changed before any auto-repair runs.