skill-scanner

To scan Claude Code skills for malicious payloads, run the skill-scanner to audit all skills in your ~/.claude/skills directory. It parses skill directories, inspects YAML frontmatter, and scans scripts to detect backdoors, prompt injections, and exfiltration attempts.

A code-scan detects malicious payloads, backdoors, prompt injections, and data exfiltration attempts embedded in skill files and bundled scripts. It identifies supply-chain risks by referencing attack-vectors guidance during local repository audits.

You can audit local skill repositories offline for supply-chain risks by running the skill-scanner. It operates across local skill directories to parse files, inspect YAML frontmatter, and scan scripts for dangerous patterns without requiring external connectivity.

YAML frontmatter inspection helps detect prompt injections by parsing skill files for malicious configurations and dangerous patterns. The scanner uses attack-vectors guidance to identify and flag suspicious metadata that could compromise system integrity.

No dependencies are required to run a malware audit on Claude Code skills. The skill-scanner operates independently to parse skill directories, inspect YAML frontmatter, and scan scripts for dangerous patterns to generate risk findings.

The best way to quarantine backdoors found in skill scripts is to use the skill-scanner's risk findings, which identify and isolate malicious payloads. It references attack-vectors guidance to safely separate compromised files during pre-install audits.