The Agent Skills Directory

[PROMPT_INJECTION]: An indirect prompt injection surface is present as the skill processes external documentation and code. 1. Ingestion points: The skill reads various project files including READMEs, API documentation, and code comments. 2. Boundary markers: There are no explicit instructions for the agent to ignore or delimit potentially malicious instructions within the data being analyzed. 3. Capability inventory: The skill writes reports to the file system and executes directory creation commands (mkdir). 4. Sanitization: No sanitization is performed on the content ingested from the analyzed files.
[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands, specifically mkdir -p, using a slug derived from user-supplied arguments. Although the skill provides instructions for the agent to sanitize this input into a URL-safe format, the capability itself represents a surface for command injection if sanitization is not properly implemented by the underlying model.

ezra-docs