The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill performs an automated global installation of the defuddle and jsdom packages using npm install -g. While jsdom is a well-known library, defuddle is an external dependency from a source not explicitly listed as a trusted vendor.
[COMMAND_EXECUTION]: The skill relies on shell command execution for both environment preparation (command -v, npm install) and its core functionality (defuddle parse).
[REMOTE_CODE_EXECUTION]: The command defuddle parse "<url>" interpolates a user-controlled URL directly into a shell execution string. This poses a risk of command injection if the agent does not strictly validate or escape the URL for shell metacharacters such as backticks or semicolons.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes content fetched from external websites.
Ingestion points: Content enters the agent's context through external URLs processed by the defuddle parse command.
Boundary markers: The skill instructions do not specify any boundary markers or delimiters to help the agent distinguish between the tool's output and potentially malicious embedded instructions.
Capability inventory: The skill has capabilities to write files to the local filesystem and execute shell commands, which could be exploited by an indirect injection attack.
Sanitization: No specific sanitization or filtering of the extracted content is mentioned before presenting the data to the user or saving it to a file.

defuddle