The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documentation includes manual installation steps that use curl to download a skill file from a non-verified GitHub repository (itsmemeworks/adhx) and save it directly into the agent's local skill directory (~/.claude/skills/). This pattern encourages users to pull unverified code from a source that does not match the stated author context.
[REMOTE_CODE_EXECUTION]: Automated scans detected a remote code execution pattern where a skill definition is downloaded from an untrusted remote URL and subsequently interpreted by the agent environment. This allows the remote source to influence the agent's behavior upon loading the skill.
[DATA_EXFILTRATION]: The skill establishes a network connection to a third-party domain (adhx.com) to process X/Twitter data. While the data sent is currently limited to URL segments (usernames and status IDs), it creates a dependency on an external, non-well-known service for handling user-provided links.
[PROMPT_INJECTION]: The skill fetches and processes social media content from a proxy API without adequate sanitization or boundary markers, making it vulnerable to indirect prompt injection attacks.
Ingestion points: Data is fetched from the adhx.com API response (SKILL.md).
Boundary markers: None; the fetched JSON fields are used directly in responses without delimiters to isolate untrusted content.
Capability inventory: The skill utilizes curl for network requests and can process long-form content for analysis.
Sanitization: There is no evidence of validation or filtering applied to the retrieved tweet or article content.

adhx