cinema-language-atlas
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: Static analysis triggers for instruction concealment were evaluated and determined to be false positives. The terms 'reveal' and 'withhold' refer to standard cinematic techniques for managing viewer information in suspenseful scenes, not attempts to hide the agent's internal operations or bypass safety guidelines.
- [COMMAND_EXECUTION]: The skill includes instructions for running local validation scripts (e.g., quick_validate.py) and search tools (ripgrep). These are presented as optional developer workflows for maintenance and use environment-specific placeholders, posing no risk to the end user.
- [EXTERNAL_DOWNLOADS]: The skill references a wide range of external educational and industry-standard domains (e.g., yale.edu, oscars.org, arri.com, kodak.com). These references are used to fetch credible film craft knowledge and are consistent with the skill's primary purpose.
- [INDIRECT_PROMPT_INJECTION]: The workflow involves browsing external sources for film examples. This inherent risk is mitigated by explicit instructions to evaluate source quality and maintain 'Misuse boundaries' for all extracted mechanisms, ensuring the agent remains focused on validated film-language controls.
Audit Metadata