context-compression

To compress context in long-running AI agent sessions, this skill uses anchored iterative summarization to reduce token usage while preserving critical information. It generates structured, delta-aware summaries to maintain relevance across multi-turn conversations exceeding context windows.

Anchored iterative summarization is a context compression technique that preserves critical information across compression cycles. It produces structured summaries with explicit sections for session intent, decisions, and next steps to maintain coherence without losing essential details.

You evaluate compressed context summaries using the probe-based evaluation workflow. This mechanism measures compression quality across recalls, artifacts, continuations, and decisions to ensure no critical details are lost during token reduction.

Yes, session compression supports artifact tracking and integrates with existing evaluation frameworks. This enables codified QA and debugging workflows when managing context length constraints in multi-turn agent conversations.

You should use structured context compression when multi-turn conversations exceed context windows and risk losing coherence. It is applicable to long-running sessions where maintaining relevance and reducing token usage are constraints.