The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructs the agent to dynamically add new capabilities using npx skills add from external URLs (e.g., https://docs.example.com/skill.md). This allows for the acquisition and execution of remote skill definitions from untrusted documentation websites.
[EXTERNAL_DOWNLOADS]: The skill relies on fetching external resources via curl and platform-specific tools from arbitrary documentation sites. This includes fetching llms.txt, llms-full.txt, and site-specific MCP server configurations.
[COMMAND_EXECUTION]: The skill provides instructions to execute shell commands such as curl and npx with parameters derived from external URLs, which can be a vector for command injection if not handled securely by the underlying platform.
[PROMPT_INJECTION]: The skill exhibits signs of metadata poisoning by claiming to be authored by "Mintlify" in the file metadata, while the provided author context identifies the publisher as "ujang19".
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to ingest and process untrusted data from external websites, creating a vulnerability to instructions embedded in third-party documentation.
Ingestion points: External content fetched from documentation URLs via curl and web fetching tools (SKILL.md).
Boundary markers: No explicit boundary markers or instructions to ignore embedded commands within the fetched documentation are present.
Capability inventory: Includes shell command execution (curl, grep), skill management (npx skills add), and MCP server integration.
Sanitization: The instructions do not specify any validation, sanitization, or filtering of the content retrieved from remote documentation sites.

doc-reader