context-compression

Use anchored iterative summarization with explicit sections for session intent, files modified, decisions, and next steps. Summarize only newly truncated content and merge it into the existing summary rather than regenerating from scratch, which prevents cumulative detail loss.

Anchored iterative summarization scores highest for coding agents because it preserves artifact trails and decision rationale. Opaque compression achieves higher ratios but sacrifices interpretability, while regenerative summaries lose details across repeated cycles.

Use probe-based evaluation instead of metrics like ROUGE. After compression, ask recall, artifact, continuation, and decision questions, then score responses across six dimensions including accuracy, artifact trail, and continuity.

Trigger compression at 70-80% context utilization as a default, or use a sliding window keeping the last N turns plus a summary. Task-boundary triggers work well when sessions have clear phase transitions like research then implementation.

General summarization cannot reliably maintain artifact trails, which benchmark as the weakest dimension across all methods. Implement a separate artifact index tracking file paths, function names, and error messages outside the summarizer.

Never compress tool definitions, function schemas, or API specs, since summarizing parameter names destroys the agent's ability to invoke tools. Also protect early session turns containing task constraints and preserve code blocks verbatim.