What problem does it solve? Long-running agent sessions generate millions of tokens of conversation history that exceed context window limits, causing agents to forget file modifications, decisions, and error details. This Skill provides compression strategies that optimize tokens-per-task rather than tokens-per-request, preventing costly re-fetching of lost context. ## Core Features & Use Cases - Anchored Iterative Summarization: Maintain structured summaries with explicit sections for session intent, files modified, decisions, and next steps, merging new content incrementally instead of regenerating. - Compression Strategy Selection: Compare anchored iterative, opaque, and regenerative approaches with concrete compression ratios and quality scores to pick the right trade-off. - Probe-Based Evaluation: Test compression quality functionally with recall, artifact, continuation, and decision probes instead of unreliable lexical metrics like ROUGE. - Use Case: A coding agent debugging a 401 authentication error across 178 messages compresses its history into a structured summary preserving the endpoint, root cause, modified files, and test status, then continues work without re-reading any files. ## Quick Start Use the context-compression skill to design a structured summarization strategy for my agent session that is exceeding its context window.
