The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests data generated by an external Grok session and fetches metadata from public x.ai bot pages, creating a potential surface for indirect injection.
Ingestion points: The agent reads a JSON array provided by the user (originating from a Grok prompt) and fetches bot page data from https://x.ai/bot/*.
Boundary markers: The skill utilizes a structured JSON schema for data ingestion as defined in the Grok prompt instructions.
Capability inventory: Capabilities include writing local data files and UI components, and executing physics simulations via a third-party library.
Sanitization: Step 2 of the skill explicitly mandates the identification and removal of private data (emails, keys, tokens, personal facts) from the ingested content, providing a strong defense against malicious data.
[EXTERNAL_DOWNLOADS]: The skill relies on the matter-js library to handle physics simulations in the generated UI.
Evidence: The physics implementation in pile.js imports the matter-js package.
Context: matter-js is a widely recognized and reputable open-source physics engine for web applications.

bots