The Agent Skills Directory

[DYNAMIC_EXECUTION]: The documentation in SKILL.md includes an example for a calculatorTool that implements tool logic using the eval() function. This pattern is highly dangerous as it allows for the execution of arbitrary JavaScript code from potentially untrusted input strings passed to the tool.
[INDIRECT_PROMPT_INJECTION]: The skill defines a framework for processing external model inputs and executing associated tools, creating a vulnerability surface for indirect instructions.
Ingestion points: The callModel function documented in SKILL.md accepts arbitrary strings and message arrays via the input parameter.
Boundary markers: The SDK provides separate instructions (system) and input (user) parameters, but the examples do not demonstrate explicit delimitation or sanitization for third-party content.
Capability inventory: The SDK enables automatic tool execution via the execute method, network requests to OpenRouter API endpoints, and programmatic management of API keys.
Sanitization: The documentation encourages the use of Zod for tool input/output validation, which provides a layer of structural and schema-based enforcement.
[EXTERNAL_DOWNLOADS]: The skill instructions and documentation reference the installation of the @openrouter/agent and @openrouter/sdk packages from the NPM registry. These packages are the core functional components provided by the skill author for model integration.

openrouter-typescript-sdk