The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions promote an insecure installation method: curl -fsSL https://raw.githubusercontent.com/xdevplatform/xurl/main/install.sh | bash. This pattern directly executes a remote script from a third-party GitHub repository not included in the trusted vendors list, providing a vector for arbitrary code execution during setup.
[EXTERNAL_DOWNLOADS]: The skill depends on the xurl CLI tool and recommends installing it via several external, unverified sources, including npm (@xdevplatform/xurl), Homebrew (xdevplatform/tap/xurl), and Go (github.com/xdevplatform/xurl). These repositories are not associated with the primary vendor (firecrawl) or known trusted organizations.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted content from the X (Twitter) platform, which can contain adversarial instructions intended to influence the agent.
Ingestion points: The skill uses tools to fetch data from the X API, such as searching for posts (xurl search), reading specific posts (xurl read), and checking mentions (xurl mentions) as described in SKILL.md.
Boundary markers: There are no boundary markers or instructions to treat the ingested JSON data as untrusted or to ignore embedded natural language instructions.
Capability inventory: The skill possesses significant capabilities, including posting content, deleting posts, following/unfollowing accounts, and sending direct messages.
Sanitization: No sanitization or filtering of the API response data is mentioned before it is returned to the agent context.

xurl