The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from X/Twitter and has capabilities to perform actions based on that content, creating a vulnerability surface.\n
Ingestion points: Tweet data is read in src/engagementManager.js (via autoLikeByKeyword and getEngagementAnalytics) and src/unlikeAllPosts.js.\n
Boundary markers: The skill documentation does not mention any delimiters or instructions to ignore commands embedded within processed tweets.\n
Capability inventory: The skill uses Puppeteer (src/engagementManager.js) to likeTweet, replyToTweet, bookmarkTweet, and hideReply.\n
Sanitization: No sanitization or validation of external tweet content is described.\n- [COMMAND_EXECUTION]: The skill requires manual execution of JavaScript code in the browser's developer console.\n
Evidence: Instructions in SKILL.md for src/unlikeAllPosts.js tell the user to "Open DevTools (F12) → Console" and "Paste the script → Enter".

engagement-interaction