The Agent Skills Directory

[SAFE]: The skill provides legitimate utility for managing context windows with OpenRouter. The network operations are directed at official OpenRouter API endpoints (openrouter.ai), which is consistent with the skill's purpose. All code examples use industry-standard libraries like openai, requests, and tiktoken. Sensitive information such as API keys is handled via environment variables, following security best practices. The external links point to the author's official domains (jeremylongshore.com, intentsolutions.io) and documentation, with no signs of typosquatting or malicious redirection.
[INDIRECT_PROMPT_INJECTION]: The skill contains functions that process untrusted external data, which is an inherent surface for indirect prompt injection given its purpose of context optimization.
Ingestion points: The document parameter in chunk_and_process() and the messages list in trim_conversation() found in SKILL.md and references/examples.md.
Boundary markers: The skill uses basic textual delimiters (e.g., Document:\n{chunk}) to separate data from instructions during processing.
Capability inventory: The skill utilizes curl for metadata lookup and the OpenAI SDK for completions across all reference scripts.
Sanitization: The skill relies on standard model interpolation and does not implement custom sanitization for the ingested text.

openrouter-context-optimization