The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill documentation recommends installing the defuddle utility globally via npm. This introduces an external dependency on a third-party package that is not part of the agent's base environment.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process content from arbitrary web pages and provide that content to the agent. This allows for potential injection attacks where a malicious website could include hidden instructions to manipulate the agent's response.
Ingestion points: Remote web content fetched from user-provided URLs via the defuddle parse command in SKILL.md.
Boundary markers: The instructions do not specify any delimiters or safety prompts to isolate the external web content from the agent's system instructions.
Capability inventory: The skill documents the ability to write extracted content to local files using the -o flag.
Sanitization: While the tool is intended to remove web clutter, there is no evidence of security-focused sanitization to prevent the processing of malicious prompts embedded in the page text.

defuddle