The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions advise the installation of the defuddle package globally via NPM (npm install -g defuddle). This introduces a dependency on an external package from a source not recognized as a trusted vendor.\n- [COMMAND_EXECUTION]: The skill utilizes the defuddle CLI to parse web content. The usage pattern defuddle parse <url> directly interpolates a URL into a shell command. If the AI agent passes unvalidated user input into this command, it could lead to shell command injection (e.g., <url> containing shell operators like ;, &&, or ||).\n- [PROMPT_INJECTION]: The skill is designed to ingest and summarize content from external web pages, which is a significant surface for indirect prompt injection.\n
Ingestion points: Web content is fetched and converted to markdown via defuddle parse <url> (SKILL.md).\n
Boundary markers: The instructions lack explicit boundary markers or directives for the agent to ignore instructions contained within the extracted content.\n
Capability inventory: The agent has the ability to execute shell commands via the defuddle CLI and potentially other tools available in the environment.\n
Sanitization: There are no specified sanitization or filtering steps to ensure that malicious instructions within the parsed web content do not influence the agent's behavior.

defuddle