Back to skill
Skillv0.1.1
ClawScan security
OpenClaw Healthcheck · ClawHub's context-aware review of the artifact, metadata, and declared behavior.
Scanner verdict
ReviewMar 15, 2026, 3:49 AM
- Verdict
- Review
- Confidence
- high
- Model
- gpt-5-mini
- Summary
- The skill's code matches its stated purpose (local OpenClaw health/security checks), but its registry metadata omits important runtime requirements and file accesses — a mismatch you should understand before running it in production.
- Guidance
- This skill appears to implement a local OpenClaw health/security audit and does not contact external endpoints, but there are two practical concerns before you run it: 1) Metadata omissions: The package registry metadata does NOT list required binaries or config paths, but the script expects system commands (openclaw, lsof, tail) and reads ~/.openclaw/openclaw.json and /tmp/openclaw/<date>.log. Confirm those commands exist and be prepared for the script to read those files. 2) Sensitive local reads: The script collects command output and file contents into a JSON 'evidence' object that may contain config values or log lines (which can include secrets or PII). Review the script first and run it in a safe/test account or sandbox (or with a temporary HOME) before running on a production host. Practical steps: - Inspect scripts/healthcheck.mjs yourself (you already have it) to verify what will be read and output. - Run the included tests in a sandboxed temp directory (npm test / node tests/test.mjs) to see behavior. - Back up ~/.openclaw/openclaw.json and any logs, or run the script with HOME pointed at a safe test directory if you want to avoid exposing real configs. - If you plan to install for repeated use, ask the maintainer to update metadata to declare required binaries (openclaw, lsof, tail) and config/log paths so the capability and the requested access are explicit. Given these mismatches between metadata and actual behavior, treat the skill as suspicious until you confirm its runtime reads and outputs are acceptable for your environment.
Review Dimensions
- Purpose & Capability
- concernThe script's actions (running `openclaw status`, lsof to inspect listeners, reading ~/.openclaw/openclaw.json, tailing /tmp/openclaw/... logs) are coherent with a local OpenClaw health/security check. However, the registry metadata declares no required binaries or config paths while the script clearly depends on system commands (openclaw, lsof, tail) and specific config/log paths. That metadata omission is an inconsistency that could lead to surprise or misuse.
- Instruction Scope
- noteSKILL.md instructs the agent to run the bundled Node script. The script reads user files (home config at ~/.openclaw/openclaw.json), inspects /tmp logs, and runs shell commands (lsof, tail, openclaw). Those actions are within the advertised scope, but they will expose potentially sensitive local content (config values, logs) into the script's JSON output (evidence). The SKILL.md does not explicitly enumerate these exact file reads/commands — the README and script do.
- Install Mechanism
- okThis is an instruction-only skill with bundled scripts; there is no network download/install step. No external packages or remote installers are fetched. Risk from installation is low, but the script will be executed locally with Node.js.
- Credentials
- concernThe skill declares no required environment variables or config paths, yet it accesses the user's HOME, reads ~/.openclaw/openclaw.json, and inspects /tmp/openclaw logs. It also expects system utilities (openclaw binary, lsof, tail). Asking for no credentials is appropriate, but the metadata should declare the config/log paths and required binaries so users know what will be read and what commands must be present.
- Persistence & Privilege
- okThe skill does not request always-on presence and does not modify other skills or system-wide settings. It runs as a one-shot diagnostic and exits (the script uses process.exit). There is no evidence of persistent background behavior or network exfiltration.