Skills Audit

Security audit + append-only logging + monitoring for OpenClaw skills (file-level diff, baseline approval, SHA-256 integrity).

This skill appears to be what it says: a local static auditor and monitor. Before enabling it, note: (1) it will read all files under your workspace/skills and store snapshots and NDJSON logs under ~/.openclaw/skills-audit—these logs can include file snippets and paths, so review their permissions; (2) the optional remote intel lookup (QianXin SafeSkill) requires you to add your token to config/intelligent.json—leave it disabled unless you trust that service and understand that only a bundle MD5 is sent, not full files; (3) the tool executes local subprocesses such as git and the included helper scripts—ensure git is installed and review the scripts if you want to audit exact behavior; (4) the skill deliberately does not auto-create cron jobs, so you must explicitly add scheduling if desired; (5) minor implementation note: the code has a default config fallback for QianXin that may differ from the shipped intelligent.json default, but the shipped config sets enabled:false—verify config/intelligent.json to be sure remote queries remain disabled. If you need to be extra cautious, run the scanner once in a restricted environment, inspect ~/.openclaw/skills-audit/logs.ndjson, and keep the QianXin token empty.

Type: OpenClaw Skill Name: skills-audit Version: 1.5.3 The bundle is a legitimate security auditing and monitoring tool for OpenClaw skills. It implements static analysis, git-based version tracking, and risk scoring using local rules and an optional, opt-in remote threat intelligence API (QianXin SafeSkill) that only transmits bundle hashes. The Python scripts (`skills_audit.py`, `skills_watch_and_notify.py`) are well-structured, use standard libraries, and perform actions consistent with the stated purpose of auditing. Crucially, the `SKILL.md` instructions contain defensive guidelines for the AI agent, explicitly prohibiting the automatic creation of cron jobs and mandating that the agent provide safe summaries of code changes rather than exfiltrating raw diffs to external channels.