image-generate
Pass
Audited by Gen Agent Trust Hub on Nov 3, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTIONDYNAMIC_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided text for image prompts and filenames which are subsequently embedded into shell command strings for tools like
agyandcodex. - Ingestion points: The
PromptandImageNamefields inreference/agy-imagen.mdandreference/codex-imagen.mdserve as entry points for untrusted data entering the agent's context. - Boundary markers: There are no documented boundary markers or instructions to the agent to ignore or escape embedded shell control characters within the prompt strings.
- Capability inventory: The agent is granted
Bashaccess and the ability to execute various scripts (recipe.py,imgfacts.py). - Sanitization: The instructions lack any requirement for sanitizing, escaping, or validating the input data before it is interpolated into shell commands.
- [COMMAND_EXECUTION]: The skill operates by invoking various CLI utilities via shell execution to manage external generators and sidecar metadata.
- Evidence includes shell command templates in
reference/agy-imagen.mdandreference/codex-imagen.mdthat use positional parameters from user input. - The documentation for the
agygenerator encourages the use of the--dangerously-skip-permissionsflag to bypass internal permission prompts. - [DYNAMIC_EXECUTION]: Command lines are constructed at runtime by concatenating base commands with parameters derived from the user's request, creating an exploitable path for potential command injection.
Audit Metadata