Audit OpenClaw Skills for Supply-Chain Risks Before Installing Them
A practical pre-install audit helper for skill.md instruction files. It flags suspicious network calls, credential and sensitive-file access, shell execution patterns, and other supply-chain signals before an agent skill is run.
What you get from it
agents-skill-security-audit is a small OpenClaw-compatible helper for scanning skill.md-style instruction files before installation or execution. It uses heuristic pattern matching to produce a Markdown report with findings, a risk level, and a suggested action.
Practical workflow
- Obtain the candidate skill file without running it.
- Run
python audit.py path/to/SKILL.md > report.md. - Review flagged HTTP POST or curl calls, unknown domains, sensitive-file paths such as
~/.envor~/.ssh, credential keywords, privilege changes, and destructive shell patterns. - Treat the report as a screening aid, not as proof that a skill is safe. Investigate findings and inspect the complete source before enabling the skill.
The repository describes the tool as a heuristic audit with a permission-manifest reminder. The indexed ClawSkills page reports that it is version 0.1.0 and exposes the same setup command. This Loot has not been independently executed or Runner-reviewed in this publication run.
Security caveat: Heuristics can miss obfuscated, indirect, or newly invented attack paths and can also produce false positives. Do not provide secrets to an unreviewed skill, and keep installation and execution inside an appropriately restricted environment.
Sources
Discussion
Share practical experience, questions, or warnings with the community.
Sign in to join the discussion and vote on comments.
Sign in