Caveat verdict
security-scanner
security-skill-scanner
A skill-scanning tool that instructs the agent to analyze SKILL.md files for suspicious patterns and report findings to the user; the network capabilities are for optional cloning from a known GitHub repo, and the content is transparent about its operation.
⚠ Flagged for review — coarse, uncorroborated signal, not a confirmed exploit. Review the config yourself before installing.
Automated static analysis — not a human review. Caveat flags capabilities, not confirmed intent, and can produce false positives. Disagree with this verdict? Use Dispute below.
Permission integrity
network_out
package_install
Findings (11)
Uses eval() — can execute arbitrary code
examples/malicious-skill/SKILL.md · code · eval(
Possible hardcoded credential
examples/clean-skill/SKILL.md · code · API_KEY="your-api-key-here
Uses exec() — may execute shell commands
examples/malicious-skill/SKILL.md · code · exec(
References SSH/GPG private keys
examples/malicious-skill/SKILL.md · code · sshKey
Accesses .ssh directory
examples/malicious-skill/SKILL.md · code · .ssh/
Reads files from sensitive system paths
examples/malicious-skill/SKILL.md · code · readFileSync('/home
References child_process — can spawn system processes
examples/malicious-skill/SKILL.md · code · child_process
Accesses sensitive environment variables
examples/clean-skill/SKILL.md · code · process.env.OPENWEATHER_API_KEY
Makes HTTP request to external URL
examples/malicious-skill/SKILL.md · code · fetch('http://
Uses XMLHttpRequest — network access
scanner.js · prose · downgraded · XMLHttpRequest
Opens WebSocket connection
scanner.js · prose · downgraded · websocket
Why the tier is capped
Execution sink present in raw bytes (Hard Floor: class B/D). Final tier capped at Caution — cannot be lifted by any downgrade, example-payload opt-in, or allowlist.
Permissions & capabilities
No declared permissions — minimal attack surface.
package_installnetwork_outnetwork_in Is this flag fair?
Thanks — recorded.