context-compression
This skill provides context-compression strategies and an evaluation toolkit for probe-based LLM judging. It includes explicit instructions to call an LLM judge (`gpt-5.2`) via `openai.chat.completions.create`, which requires outbound network calls.