The Agent Skills Directory

[SAFE]: The skill is a specialized defensive utility for identifying supply chain attacks. Its behavior is consistent with its stated purpose of forensics and auditing.- [EXTERNAL_DOWNLOADS]: The automated scan flagged several URLs (e.g., sfrclak.com, npm-cache.com, git-tanstack.com, api.masscan.cloud) as malicious. These are correctly identified as Command and Control (C2) or typosquatting domains associated with supply chain malware. They are included in the skill's iocs.json file as search patterns (Indicators of Compromise) for detection purposes only. The skill does not attempt to connect to or download from these URLs.- [COMMAND_EXECUTION]: The skill executes a local bash script (scripts/scan.sh) to perform its audit. The script uses standard system utilities (find, grep, jq, shasum, git log) in a read-only manner. It operates within a temporary directory and does not modify user files, aligning with safe operational practices for forensics tools.- [DATA_EXFILTRATION]: The skill includes 'Hard Constraints' that explicitly forbid accessing sensitive credential files (e.g., ~/.ssh/, ~/.aws/, .env) or making network calls. The logic in scripts/scan.sh adheres to these constraints, focusing only on project metadata and known malware artifacts.- [INDIRECT_PROMPT_INJECTION]: While the skill processes untrusted local data (like package.json or node_modules paths), it mitigates potential context hijacking by using structured markdown output and escaping characters (such as pipes) that could interfere with the agent's interpretation of the report.

supply-chain-audit