The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs users to clone a repository from an external source (https://github.com/EastSword/skill-dfyx_code_security_review.git) which is not associated with the primary vendor or a known trusted organization.
[DATA_EXFILTRATION]: The SecretFinder component is programmed to locate and read highly sensitive information, including AWS credentials, private keys, and environment files (.env). While intended for security auditing, this capability constitutes a data exposure risk if the agent is directed to unintended paths.
[COMMAND_EXECUTION]: The PatternScanner and SecretFinder scripts use filesystem operations (os.walk) to traverse directories and read the content of all files in the project path, providing broad access to the execution environment's data.
[PROMPT_INJECTION]: The skill acts as an indirect prompt injection surface as it ingests and processes untrusted code from external projects without sanitization or boundary markers.
Ingestion points: Project files accessed via os.walk in pattern_scanner.py and scripts/code_scan.py.
Boundary markers: Absent; file contents are read directly into analysis buffers and then summarized for the user.
Capability inventory: Full file read capabilities across the project directory and automated report generation.
Sanitization: Absent; content is analyzed using regex and AST parsing, which can be bypassed or manipulated by specially crafted content in the audited files.

dfyx-code-security-audit