The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes various project scripts using the bun runtime, including bun test, bun run render-element-previews, and bun run upload-element-preview. These commands operate on local files and repository state, including code provided in external pull requests.
[INDIRECT_PROMPT_INJECTION]: The skill processes code and assets from external contributor pull requests, creating a surface for indirect prompt injection.
Ingestion points: The agent is instructed to identify and process the <category>/<slug> and associated implementation code from the pull request branch (e.g., packages/docs/static/elements).
Boundary markers: The skill explicitly includes instructions to "Stop and wait for explicit approval of the final source and both committed assets before uploading. Do not infer approval from the agent's own inspection."
Capability inventory: The skill utilizes shell execution via bun, performs network operations to upload assets to R2/Cloudflare via vendor-specific scripts, and modifies the local filesystem before performing git commits and pushes.
Sanitization: The skill relies on manual maintainer visual verification rather than automated sanitization of the processed PR content.

accept-element