The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection during its analysis phase. It ingests data from implementation files located at user-defined paths without employing boundary markers or explicit instructions to ignore embedded directives. Given that the skill is granted capabilities including filesystem writes and shell execution (Bash), malicious content within the analyzed files (such as comments) could theoretically attempt to influence the agent's actions.
Ingestion points: Files are read from the <path> argument using tools like Read, Glob, and Grep in Phase 2.
Boundary markers: No delimiters or isolation instructions are present to separate untrusted implementation data from the agent's system instructions.
Capability inventory: The skill uses Bash, Write, and Edit tools.
Sanitization: Content read from the filesystem is not sanitized or escaped before processing.

reverse-document