grant-figure-qa
Pass
Audited by Gen Agent Trust Hub on Sep 21, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill instructions and procedure file (
references/figure-qa-procedure.md) direct the agent to execute shell commands such asfindandidentifyto locate and analyze image files within the local proposal directory. - [DYNAMIC_EXECUTION]: The skill utilizes
python3 -cto run a Python snippet that imports the Pillow library to extract image metadata like resolution and DPI. - [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes untrusted data (filenames and file contents) from the user's directory. If a filename is maliciously crafted (e.g., containing shell metacharacters), it could lead to command injection if the agent blindly interpolates the name into the provided command templates.
- Ingestion points: Filenames and figure files located within the proposal directory, as specified in
references/figure-qa-procedure.md. - Boundary markers: The instructions do not define boundary markers or 'ignore' directives for the data being processed.
- Capability inventory: The skill has the capability to search the filesystem, execute shell commands, and run Python code.
- Sanitization: The procedure lacks explicit sanitization or escaping requirements for the filenames used in the command-line templates.
Audit Metadata