game-sprite-pipeline
Fail
Audited by Gen Agent Trust Hub on Aug 2, 2026
Risk Level: HIGHPROMPT_INJECTIONCOMMAND_EXECUTIONDATA_EXFILTRATIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The character generation logic is vulnerable to prompt injection attacks where an attacker can supply instructions instead of an image description to hijack the backend agent's execution flow. This occurs in
sprite_generate.pywhen constructing prompts for thecodexCLI. - [COMMAND_EXECUTION]: The pipeline invokes the
codexCLI with the--dangerously-bypass-approvals-and-sandboxflag, which instructs the backend agent to execute shell commands without human approval or security sandboxing, enabling arbitrary command execution. - [DATA_EXFILTRATION]: Sensitive system data such as SSH keys and cloud credentials can be exfiltrated through the command execution vector, as
sprite_generate.pycaptures and logs process output. - [COMMAND_EXECUTION]: Integration scripts such as
road_to_aew_integration.pyexecute shell commands (npm run) in user-controlled directories, creating a risk of malicious script execution. - [EXTERNAL_DOWNLOADS]: The skill recommends and relies on external third-party dependencies such as the
@openai/codexCLI and therembgPython package.
Recommendations
- AI detected serious security threats
Audit Metadata