shot-planning

Pass

Audited by Gen Agent Trust Hub on Nov 4, 2026

Risk Level: SAFE
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided content such as scripts and scene descriptions. It proactively mitigates the risk of indirect prompt injection through multiple 'Hard Rules' (e.g., Rule 8 in references/zerofield-shot-beats.md) which explicitly command the agent to treat story text and tool outputs strictly as data and ignore any embedded instructions.
  • [COMMAND_EXECUTION]: The skill utilizes a 'proposals-only' architecture. Tools like propose_shot_prompt and propose_previs_ops allow the agent to suggest changes, but it is strictly forbidden from adopting its own proposals or writing directly to the project without human intervention.
  • [DYNAMIC_EXECUTION]: In the previs blocking stage, the skill explicitly prohibits the use of code (Python, JavaScript, etc.), restricting the agent to a closed vocabulary of staging operations that are processed by a deterministic solver, thereby preventing arbitrary code execution.
  • [TIME_DELAYED_CONDITIONAL]: The skill incorporates human-in-the-loop 'Gates' at every stage of the planning process. This ensures that no automated actions are taken, particularly regarding resource spending, which requires an explicit 'Gate 4' approval from the user after a cost estimate is presented.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 4, 2026, 08:11 PM