The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from user-supplied briefs and external reference URLs to guide code generation and execution.
Ingestion points: The <brief> argument and content from external URLs provided via the --ref flag in SKILL.md are used to define the generated code's behavior.
Boundary markers: Absent; the instructions do not include specific delimiters or warnings for the agent to ignore potentially malicious instructions embedded within external reference content or user briefs.
Capability inventory: The skill is configured with Bash and Write tools, which are used to generate project files and execute local environment commands such as npm install and python3 servers as described in references/run.md.
Sanitization: Absent; the skill directs the agent to "build bespoke" implementations directly from the provided brief without explicit steps to sanitize or filter the input data for prompt injection patterns.
[EXTERNAL_DOWNLOADS]: The skill references and downloads project dependencies from well-known and established services.
Evidence: references/stack.md specifies the use of cdnjs.cloudflare.com, cdn.jsdelivr.net, and unpkg.com for loading libraries such as Three.js, GSAP, and Lenis.
Evidence: references/run.md indicates that npm install is used to manage dependencies for React-based project targets.
[COMMAND_EXECUTION]: The skill utilizes shell commands to manage development tasks and serve generated content.
Evidence: references/run.md contains instructions for serving HTML output via python3 -m http.server and managing Node.js project lifecycles with npm run dev.

motion-site