The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and follow instructions (the 'contract') returned by the compose_block_brief tool. This creates a surface where the agent is instructed to adopt external system and prompt instructions 'verbatim'. While this is the intended workflow for the 'BYO-brain' feature, it relies on the integrity of the data sources (theme tokens, frame designs) used by that tool. The risk is mitigated by the described validation step in the apply_block tool which checks for scripts and animation constraints.
[DYNAMIC_EXECUTION]: The core functionality involves the agent generating HTML markup and JavaScript animation timelines at runtime. The skill explicitly mentions that the apply_block tool performs validation (linting) to ensure the code is scoped, contains no scripts, and uses deterministic animations, which serves as a security control for the dynamic content.

compose-blocks