What problem does it solve?
Large agent sessions produce enormous conversation histories, blowing through context windows and increasing latency. This skill provides strategies to compress history while preserving essential intent, decisions, and state.
Core Features & Use Cases
- Anchored Iterative Summarization: Maintain structured, persistent summaries with explicit sections for session intent, file modifications, decisions, and next steps. When compression triggers, summarize only new content and merge it into the existing structure to avoid losing critical details.
- Opaque Compression: Generate highly compact representations that maximize token savings at the cost of readability, suitable for long-running sessions with high re-fetch costs.
- Regenerative Full Summary: Produce detailed, human-readable summaries on each compression cycle to facilitate quick reviews and audits.
- Artifact Tracking: Capture and track which files were read, modified, or created to preserve an accurate history trail across compressions.
- Evaluation-Probe Ready: Include mechanisms to test recall, artifact integrity, and continuation capability after compression.
Quick Start
Use the context-compression skill to start structuring a session memory and enable incremental compression when token usage grows.
