The Agent Skills Directory

[SAFE]: The skill incorporates a robust internal version control mechanism. Rollback manifests located in the versions/ directory utilize SHA256 hashes to verify the integrity and authenticity of instructions and reference files, ensuring that the agent uses only authorized snapshots of the skill.
[PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it is designed to ingest and process user-provided scripts and production materials.
Ingestion points: SKILL.md instructions for reading full scripts and assets.
Boundary markers: Structurally absent, but the skill mandates a conceptual distinction between script facts and agent interpretations.
Capability inventory: The skill is restricted to generating text-based storyboard tables and video prompts; it lacks file-system modification or network exfiltration capabilities.
Sanitization: No explicit data sanitization is mentioned, but the rigorous requirement to maintain world-state consistency and physical logic acts as a significant barrier to simple instructional overrides.
[SAFE]: The skill adheres to strict operational boundaries by explicitly forbidding the invocation of invalidated or experimental versions (5.5 and 6.0) and mandating that all outputs comply with the verified production contracts in the references/ folder.

ai-storyboard-director