The Agent Skills Directory

[SAFE]: The skill is primarily instructional, providing high-quality code examples and architectural guidance for integrating OpenRouter into applications.
[EXTERNAL_DOWNLOADS]: The skill recommends the use of official, well-known libraries for its implementation.
Evidence: Recommends openai (Node.js/Python) and tiktoken (Node.js) for API interaction and token counting.
[INDIRECT_PROMPT_INJECTION]: The skill defines patterns for processing untrusted user data within AI prompts, which introduces an indirect injection attack surface. However, it explicitly mitigates this through specialized documentation.
Ingestion points: User-provided prompts processed in functions like handleSubmit(prompt) and chatWithFunctions() in SKILL.md.
Boundary markers: Includes recommendations for using system prompts to enforce constraints and separate developer instructions from user content.
Capability inventory: Network operations via the OpenAI SDK and standard fetch calls.
Sanitization: The "Best Practices" section explicitly mandates validating user inputs before sending them to the API.
[DATA_EXFILTRATION]: While the skill manages API keys, it follows best practices by instructing users to use environment variables and server-side proxies, explicitly warning against exposing keys in client-side code.
Evidence: "Never expose API keys in frontend code
Use server-side proxies" is listed under Best Practices in README.md and SKILL.md.

openrouter