The Agent Skills Directory

[SAFE]: The skill executes the official Grok CLI tool using node:child_process.execFile with strictly validated arguments. This approach prevents shell injection and ensures that only intended search parameters (query, handles, dates) are processed.
[SAFE]: The helper script (search-x.mjs) performs a security audit of the local Grok configuration before execution. It rejects any custom model overrides, unofficial endpoints, or hardcoded API keys, ensuring that search data is only sent to the official xAI infrastructure.
[SAFE]: The Grok process is launched with a highly restrictive sandbox configuration. It explicitly disables capabilities such as terminal access, file system writes, and generic web fetching to prevent the search tool from being abused for unintended actions.
[SAFE]: The skill provides clear guidance to the AI agent to mitigate indirect prompt injection. It instructs the agent to treat all content retrieved from X as untrusted data and to disregard any commands or instructions found within the posts.
[SAFE]: Sensitive authentication data is protected by explicit rules in the skill instructions that forbid the agent from accessing or outputting tokens, session IDs, or authorization headers.

x-search