What problem does it solve? Long-running agent sessions generate millions of tokens of conversation history that exceed context windows, and naive compression loses critical details like file paths, error messages, and decision rationale, forcing costly re-exploration. ## Core Features & Use Cases - Anchored Iterative Summarization: Maintain persistent structured summaries with explicit sections for session intent, file modifications, decisions, and next steps, merging new spans incrementally instead of regenerating from scratch. - Probe-Based Evaluation: Generate recall, artifact, continuation, and decision probes from conversation history, then score responses across six dimensions (accuracy, context awareness, artifact trail, completeness, continuity, instruction following) using an LLM judge rubric. - Compression Method Selection: Choose between anchored iterative, opaque, and regenerative summarization based on session length, file-tracking needs, and re-fetching costs, with benchmark data comparing compression ratios and quality scores. - Use Case: A coding agent debugging a 401 error across 178 messages hits the context limit; the skill compresses history into a structured summary preserving the root cause, modified files, and failing tests, then validates quality with probes before discarding the original history. ## Quick Start Ask the agent to compress the current conversation history into a structured summary with sections for session intent, files modified, decisions, and next steps.
