The Agent Skills Directory

[SAFE]: The skill implements a robust 'Delivery Gate Contract' and 'Hard Gates' (documented in references/operations.md and references/qc-gates.md) that mandate multiple levels of verification before any production output is considered usable. This approach prevents the accidental or malicious generation of unauthorized content.
[SAFE]: A clear 'Source Discipline' policy is established, requiring all production rules to be derived from a compiled, verified knowledge cache. The skill explicitly requires an audit trail (00_knowledge_read_audit.md) for any raw external data search, mitigating the risk of indirect prompt injection.
[SAFE]: Credential security is prioritized; the instructions in operations.md and product/combat_canvas_integration.md explicitly forbid the storage of raw API keys in project files or chat history, recommending secure host-side environment variables instead.
[SAFE]: The skill defines clear boundaries in references/boundaries.md, distinguishing the pipeline's purpose from general software development or high-end promos, ensuring that the agent remains within its intended scope and avoids dangerous tool-chaining.
[SAFE]: All mentioned external service providers (e.g., Seedance, Runway, Vidu) are recognized industry services, and the skill provides specific, transparent strategies for interacting with them via structured compiler outputs rather than raw code execution.

short-drama-pipeline