invokeai-image-gen
Pass
Audited by Gen Agent Trust Hub on Nov 11, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill includes a Python script (
scripts/generate.py) designed to be run in a shell environment. The script can write files to the local disk based on provided arguments, which could be exploited for unauthorized file modification if the agent's input is not strictly controlled.- [EXTERNAL_DOWNLOADS]: The script usesurllib.requestto fetch image data from an external API, introducing a mechanism for downloading remote content onto the local system.- [DATA_EXFILTRATION]: The skill transmits user prompts and environment-sourced authentication tokens to a remote API endpoint. Directing this traffic to an untrusted URL would result in the exfiltration of sensitive data.- [INDIRECT_PROMPT_INJECTION]: The skill handles untrusted user input and external API data, forming an injection surface. - Ingestion points: User-provided prompts and negative prompts, along with data returned from the InvokeAI API.
- Boundary markers: No explicit delimiters are used to isolate untrusted content within the agent's context.
- Capability inventory: Network connectivity and file system write operations are present in
scripts/generate.py. - Sanitization: While API payloads use
json.dumps()for basic escaping, there is no validation or sanitization of content received from the remote API.
Audit Metadata