The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external sources, including social media post histories, search results, and thread content, to inform its deduplication logic and engagement activities. The instructions do not define boundary markers or sanitization procedures for this external content before it is interpolated into prompts for content generation. A malicious post or thread could theoretically include instructions designed to influence the agent's behavior during these steps.
Ingestion points: xurl timeline and xurl search results, Discord and Telegram message history reading, and web scraping of profiles and threads using Puppeteer.
Boundary markers: None are specified for the processing of external social media content.
Capability inventory: The skill can publish content via xurl and platform APIs, execute UI actions through peekaboo (click, type, hotkey), and run system commands like ffmpeg.
Sanitization: The skill lacks explicit sanitization or filtering logic for external data prior to its use in generating drafts or replies.
[COMMAND_EXECUTION]: The skill uses several command-line tools for its core functionality, including xurl for X/Twitter interactions, ffmpeg for video processing, and peekaboo for macOS UI automation. It also makes use of curl to communicate with platform APIs (Telegram, Discord, Instagram). These tools provide the agent with significant system-level capabilities.
[DYNAMIC_EXECUTION]: The technical references for browser automation describe the use of page.evaluate and Runtime.callFunctionOn (via the Chrome DevTools Protocol) to execute JavaScript within the context of a browser session. Additionally, the skill uses python3 -c to generate random numeric identifiers for API uploads. These patterns involve the generation and execution of code at runtime to facilitate automation.

social-media-posting