The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data by searching and reading files within a workspace using tools like rg, sed, and cat. This creates an attack surface where malicious instructions embedded in code (e.g., in comments or string literals) could attempt to influence the agent or the caller receiving the analysis.
Ingestion points: Files throughout the workspace root.
Boundary markers: The instructions do not define specific delimiters or warnings to treat ingested code content as data rather than instructions.
Capability inventory: The skill utilizes shell_command for searching and reading files. It explicitly forbids modification, builds, or network-related operations, which significantly limits the impact of potential injections.
Sanitization: There are no instructions for sanitizing or escaping code content before reporting it to the caller.

searching-code