The Agent Skills Directory

[NO_CODE]: The skill frequently references src/grokIntegration.js and src/threadComposer.js as the core logic providers, but these files are missing from the provided content, making the actual implementation unverifiable.
[COMMAND_EXECUTION]: The documentation instructs users to "Paste the script -> Enter" into the browser DevTools console on x.com. This pattern of executing unverified JavaScript in a logged-in session is a high-risk vector for session hijacking (Self-XSS) and unauthorized account manipulation.
[PROMPT_INJECTION]: The skill defines prompt templates that interpolate untrusted data (e.g., {niche}, {topic}, {tweet_text}) directly into the AI context via XActions.ask(). This creates an indirect prompt injection surface.
Ingestion points: Variable data from tweets or user input is passed to Grok.
Boundary markers: No delimiters or safety instructions are used to separate user data from the system prompt.
Capability inventory: The skill possesses DOM scraping and automated submission capabilities.
Sanitization: No escaping or validation of the interpolated strings is evident.
[DATA_EXFILTRATION]: The functions XActions.scrapeResponse() and XActions.export() are designed to read and aggregate data from the X.com DOM, including AI-generated content and potentially private session data, which could be exfiltrated if the missing scripts contain network calls.
[EXTERNAL_DOWNLOADS]: The skill mentions an external "XActions MCP server" and integration with OpenRouter, which introduces third-party dependencies and the requirement for external API keys (OPENROUTER_API_KEY).

grok-ai