The Agent Skills Directory

[DATA_EXFILTRATION]: The skill explicitly warns against sending sensitive data to the remote service. It mandates getting authorization before disclosing private documents, internal URLs, or personal data to markdown.new. It specifically cautions that the service's retention policy is unknown.
[COMMAND_EXECUTION]: The skill uses curl commands to interact with the markdown.new API. These are used for legitimate service interaction (GET/POST requests for conversion). The instructions include safety warnings about validating input before sending it to these commands.
[INDIRECT_PROMPT_INJECTION]: The skill includes a dedicated 'Data boundary' section that explicitly instructs the agent to treat converted Markdown, metadata, and error bodies as untrusted content and never as instructions. This is a best-practice defense against instructions that might be embedded in the crawled/converted web content.
[EXTERNAL_DOWNLOADS]: The skill interacts with https://markdown.new/, which is a well-known utility service for Markdown conversion. These interactions are documented neutrally and are central to the skill's primary purpose. No automated or hidden downloads of executable code were detected.
[CREDENTIALS_UNSAFE]: The skill correctly advises against forwarding source-site credentials, cookies, or authorization headers to the service, noting that it operates as a public client.

markdown-new