The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data in the form of user-supplied prompts, images, videos, and audio clips which are then used as inputs for generative models.
Ingestion points: Data enters the agent context through media URL fields (video_urls, image_urls, audio_urls) and text prompts in generation tools described in SKILL.md and model-specific references.
Boundary markers: The skill employs explicit reference tokens (e.g., @Image1, <IMAGE_REF_0>, Image 1) and prompt formulas to separate instructions from external assets.
Capability inventory: The skill utilizes powerful tools for media generation (generate_video, generate_image, generate_voiceover) and file manipulation (assemble_film, add_audio_to_video, merge_media) across its entire instruction set.
Sanitization: Instructions require quoting exact dialogue and assigning specific roles to references, reducing the risk that the model will interpret embedded instructions as primary commands.
[DYNAMIC_EXECUTION]: The skill facilitates the use of HTML-based rendering for specialized visual content.
Evidence: SKILL.md mentions the use of creativeclaw-render-html for motion graphics, kinetic typography, and code-driven video rendering.
Context: This capability is restricted to specific user requests for animated explainers and is handled as a creative rendering tool rather than arbitrary code execution.

creativeclaw-build-film