ai-video-iteration-doctor
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill provides logic and documentation for debugging generative media outputs. It does not perform network operations, access the local filesystem, or execute shell commands. All referenced tools appear to be internal prompts or specialized agents within the same ecosystem.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted data in the form of user feedback and AI-generated outputs to generate prompt patches.
- Ingestion points: User feedback descriptions (e.g., "result is wrong") and generated videos/images analyzed in SKILL.md.
- Boundary markers: The skill uses a defined "Output Shape" and the "Retry Patch Template" in the diagnosis map to separate internal analysis from the proposed prompt patch.
- Capability inventory: The skill's capabilities are restricted to calling other diagnostic or formatting skills (e.g., $jimeng-sd2-prompting, $visual-style-aesthetic-direction). No evidence of subprocess calls, file writing, or network transmission was found.
- Sanitization: The skill does not explicitly perform sanitization or escaping of user-provided feedback before it is incorporated into the iteration prompts.
Audit Metadata