The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to install the markit-ai package from the NPM registry using commands like npm install -g markit-ai.
[REMOTE_CODE_EXECUTION]: The skill's plugin architecture allows downloading and executing code from arbitrary remote sources. For example, the command markit plugin install git:github.com/user/markit-plugin-ocr fetches and installs code from a Git repository.
[DYNAMIC_EXECUTION]: The tool supports loading and executing custom converter plugins at runtime from local files (e.g., markit plugin install ./my-plugin.ts) and external packages, which allows for dynamic code execution within the agent's environment.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which could contain malicious instructions designed to manipulate the AI agent.
Ingestion points: Untrusted data enters the context via markit <url> and markit <file> commands (supporting PDF, Word, HTML, etc.), as well as convertUrl() and convertFile() SDK methods.
Boundary markers: None are specified in the provided instructions to delimit untrusted content or warn the agent about embedded instructions.
Capability inventory: The skill has the capability to perform network operations (calling LLM APIs like OpenAI and Anthropic), write files to disk (using the -o flag), and install/execute plugins.
Sanitization: There is no mention of sanitization, validation, or filtering of the external content before it is processed by the LLM for vision or transcription tasks.

markit-markdown-converter