The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to install the defuddle package from the NPM registry via the global install command npm install -g defuddle.
[COMMAND_EXECUTION]: The skill uses the defuddle CLI tool to process external URLs, which involve executing a local subprocess.
[INDIRECT_PROMPT_INJECTION]: Processing external web content creates an attack surface for instructions embedded in data.
Ingestion points: Remote content is fetched and parsed via the defuddle parse <url> command defined in SKILL.md.
Boundary markers: There are no explicit delimiters or instructions provided to the agent to distinguish the parsed content from trusted instructions or to warn against executing commands found within the data.
Capability inventory: The skill is limited to running the defuddle CLI; it does not request file-writing capabilities or access to sensitive credentials in the provided instructions.
Sanitization: Although the tool converts HTML to markdown, which removes some script tags, it does not specifically sanitize the extracted text to prevent the LLM from obeying natural language instructions found in the output.

defuddle