The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill attempts to install a package from the NPM registry using the command npm install -g defuddle. This introduces a dependency on a third-party tool that is not part of a known trusted ecosystem, which can pose a supply chain risk if the package is compromised or malicious.
[COMMAND_EXECUTION]: The instructions direct the agent to execute several shell commands, including environment checks (which defuddle), global package installation, and execution of the defuddle binary with various flags. Specifically, it uses the tool's capability to write to the file system using the -o flag.
[PROMPT_INJECTION]: The skill is designed to ingest data from arbitrary external URLs and process them into markdown for the agent's consumption. This creates a surface for indirect prompt injection attacks.
Ingestion points: Web content is fetched and converted via the defuddle parse <url> command (SKILL.md).
Boundary markers: The skill lacks explicit instructions for the agent to wrap the scraped content in delimiters or to ignore any instructions found within the processed text.
Capability inventory: The agent has the ability to execute shell commands, install software, and write files to the local disk (SKILL.md).
Sanitization: There is no evidence of sanitization or filtering of the extracted text to remove potential injection vectors before it is presented to the agent.

defuddle