The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill instructs the agent to write a JavaScript validation script (.motion/scripts/beats-check.mjs) to the filesystem and then execute it using Node.js.
Evidence: references/templates/beats.md contains the source code for the script and the command to run it.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from the user's codebase (tokens, screens) and external content (reviews, tickets, comments) to generate film briefs. This data is used to define the film's structure and can influence the content of generated scripts.
Ingestion points: Codebase exploration for product tokens as described in SKILL.md; interpolation of user reviews or tickets in references/templates/brief.md.
Boundary markers: There are no explicit delimiters or boundary instructions for the interpolation of external data.
Capability inventory: The agent is capable of writing files and executing shell commands such as node, bun, and ffmpeg.
Sanitization: The skill mentions a safety feature where writes are scanned against a banned term list located at ~/.fuse-harness/motion/banned.txt and redacted if matches are found.
[COMMAND_EXECUTION]: The pipeline relies on shell commands for rendering, validation, and media inspection tasks.
Evidence: The instructions refer to the use of node, bun, ffprobe, and ffmpeg for pipeline operations.

motion-method