The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions and metadata specify the execution of local shell commands (npm install, node scripts/validate.js) to validate generated output bundles against the defined security axioms. This capability is used exclusively for local integrity and compliance checks.
[EXTERNAL_DOWNLOADS]: The skill uses npm to install the ajv package from the well-known npmjs.org registry. This is a standard dependency for performing JSON Schema validation within the provided validation script.
[PROMPT_INJECTION]: The skill manages a potential indirect prompt injection surface by ingesting untrusted user data (variables_used) and interpolating it into prompts for high-capability video models. However, this is mitigated by robust safety mechanisms:
Ingestion points: User-supplied values for template variables in Phase 1.
Boundary markers: Use of strict template interpolation placeholders {key} and model-specific wrapping structures.
Capability inventory: Generation of prompts for external models and execution of a local Node.js validation script.
Sanitization: Automated scanning for banned words (cinematic, flawless, etc.) via scripts/validate.js, literal rendering enforcement (Axiom 1), and scene-lighting word blacklists (Axiom 3) defined in schema.json.

100x-prompt-compose