layered-history-compression-reuse

Layered history compression reduces token usage by preserving recent turns verbatim, truncating middle-age tool results, and eliding very old interactions to maintain conversational context. This keeps conversations performant while retaining essential recent data.

Layered history compression runs tool-pair safety after compression to automatically repair orphaned tool_use and tool_result pairs. This defensive management ensures API integrity is maintained even when older tool results are truncated or elided.

The best way to truncate middle-age chat history is applying a layered tiering approach that keeps recent content verbatim, shrinks middle-age tool results, and fully elides the oldest interactions to optimize token consumption.

Yes, layered history compression integrates directly with snipCompact and the SnipResult contract. This extensible architecture supports env-gate logic and allows for safe handling of non-text blocks while enabling future tier extensions.

You need layered tiering when ongoing conversations experience high token usage and performance degradation. It is specifically required in tool-integrated conversations where preserving tool_use and tool_result pair integrity is critical.

Eliding older interactions can break tool API integrity by orphaning tool_use and tool_result pairs. Layered history compression solves this by running tool-pair safety after compression to repair the broken pairs and maintain API validity.