big-idea-video
Pass
Audited by Gen Agent Trust Hub on Oct 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in the form of cut transcripts and video examples which guide the agent's code modification and command execution steps.\n
- Ingestion points: Step 1 (Transcript with word timings, noun of each beat, and multiple media examples).\n
- Boundary markers: None; the instructions do not include delimiters or warnings to ignore potentially malicious instructions embedded in the user-supplied data.\n
- Capability inventory: File-write (modifying video.js in Step 3) and Subprocess call (executing node render.mjs in Step 2).\n
- Sanitization: No sanitization or schema validation is applied to the timing data or transcript text before they are passed to the rendering engine or script editor.\n- [DYNAMIC_EXECUTION]: The skill instructs the agent to dynamically modify a local JavaScript file (video.js) based on parameters derived from user input, which constitutes runtime code generation.\n- [COMMAND_EXECUTION]: The skill executes a local node process (node render.mjs) using timing values extracted from the user-provided transcript as arguments for the --stills flag.
Audit Metadata