generate-image
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill constructs a natural language instruction for the
codex exectool by interpolating user-provided data. This creates a surface where malicious user input could potentially instruct the sub-agent to perform unauthorized actions.\n - Ingestion points: The prompt argument in
scripts/generate-image.sh, which originates from user input processed by the agent.\n - Boundary markers: The
$PROMPTvariable is placed directly into thecodex execstring without delimiters or instructions to ignore embedded commands.\n - Capability inventory: The
codex exectool is grantedworkspace-writecapabilities, allowing it to modify files.\n - Sanitization: No sanitization or filtering is performed on the prompt content before interpolation.\n- [COMMAND_EXECUTION]: The bash script executes the
codexCLI tool to perform its primary function.\n - Evidence:
scripts/generate-image.shcallscodex login statusandcodex execwith specified arguments.
Audit Metadata