The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides a surface for indirect prompt injection by interpolating untrusted data into prompt templates.
Ingestion points: Data enters via the variables argument in the raison.render(id, variables) method, as seen in SKILL.md and references/examples.md.
Boundary markers: None are enforced or recommended by the SDK instructions to isolate template variables from prompt instructions.
Capability inventory: The rendered prompts are intended for use with powerful LLM models (Claude, GPT-4) which may have tool-use capabilities.
Sanitization: The SDK documentation notes that noEscape: true is used, disabling standard character escaping, and no specific sanitization steps are recommended for interpolated content.
[DYNAMIC_EXECUTION]: The Raison.registerHelper static method allows developers to register custom JavaScript functions that are executed at runtime during the template rendering process. This enables dynamic code execution within the prompt rendering pipeline.
[EXTERNAL_DOWNLOADS]: The skill documentation instructs the user to install the raison package from the NPM registry, along with other integration libraries like openai and @anthropic-ai/sdk.
[COMMAND_EXECUTION]: The skill documentation includes standard shell commands for package installation using npm install.

raison-sdk