context-compression

Compress context in long-running AI agent sessions by applying anchored iterative summarization to prune history while preserving critical decisions, file changes, and state for reliable continuation.

Anchored iterative summarization is a context compression strategy that uses explicit sections to prevent information drift while reducing token bloat in persistent AI agent workflows.

The best way to preserve state is to compress history into structured, searchable summaries that retain critical decisions and file changes using an evaluation framework with probes to measure information preservation and task continuity.

Yes, you can use multiple context compression strategies including Anchored Iterative, Regenerative, and Opaque approaches, each with specific trade-off guidance for scenarios like debugging and design reviews.

Evaluate information preservation using a built-in evaluation framework that employs probes to measure how well critical decisions, file changes, and state are retained for task continuity.

No specific dependencies are required to manage token bloat; the context compression approach uses a local state directory and on-demand references with a SKILL.md frontmatter containing a name and description.