omx
Fail
Audited by Gen Agent Trust Hub on Mar 11, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill recommends the
--madmaxflag, which maps to--dangerously-bypass-approvals-and-sandbox, explicitly disabling safety sandboxes and approval requirements. - [EXTERNAL_DOWNLOADS]: Downloads and installs the third-party
oh-my-codexpackage from the npm registry. - [REMOTE_CODE_EXECUTION]: Features a lifecycle hook system that executes JavaScript files from the
.omx/hooks/directory during execution. - [PROMPT_INJECTION]: Permits overriding system instructions via environment variables or local
AGENTS.mdfiles. This exposes a vulnerability surface where untrusted data from task descriptions (ingestion points in workflow skills) can be processed using powerful capabilities like Bash and hook execution without explicit boundary markers or sanitization. - [COMMAND_EXECUTION]: Instructs the user to install a global npm package and run setup and diagnostic scripts.
Recommendations
- AI detected serious security threats
Audit Metadata