The Agent Skills Directory

[SAFE]: The skill is designed as a security auditing tool. Its instructions and logic are focused on identifying risks in other AI agent skills, such as credential theft or malicious command execution.
[COMMAND_EXECUTION]: The skill uses grep and file reading tools to perform its audit functions. This is consistent with its intended purpose as a static analysis tool.
[DATA_EXFILTRATION]: The skill instructions include patterns to search for sensitive files (e.g., .ssh, .aws/credentials) and external URLs. This behavior is diagnostic; there is no evidence of the skill collecting this data for unauthorized external transmission.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests and analyzes untrusted data from external files.
Ingestion points: Reads SKILL.md and various source code files (e.g., .py, .js, .sh) from user-specified directories.
Boundary markers: Absent. The skill does not define specific delimiters or instructions to ignore embedded prompts in the analyzed content.
Capability inventory: The agent can list files, read file contents, and execute grep commands.
Sanitization: The skill relies on AI semantic analysis to differentiate between benign and malicious code but lacks programmatic sanitization of the input data.

scan-skill