idea-refine
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to use
Glob,Grep, andReadto scan the project's codebase for architectural and prior art context. This allows external data from the project to enter the agent's prompt context. - Ingestion points: Local codebase files and user-supplied idea descriptions via
$ARGUMENTS. - Capability inventory: Filesystem read (Glob/Read/Grep), file write (saving one-pagers), and local script execution.
- Boundary markers: No delimiters or explicit instructions are provided to the agent to treat file content as untrusted data.
- Sanitization: No validation or sanitization of the content read from files is performed.
- [COMMAND_EXECUTION]: The skill provides a shell script
scripts/idea-refine.shintended to initialize a directory for saving ideas. The script is static, performs benign filesystem operations (mkdir), and is presented as an optional manual step for the user.
Audit Metadata