strategic-compact

Manual context compaction at logical task boundaries preserves critical information during long AI sessions. By compacting after a research phase instead of relying on arbitrary auto-compaction, you retain essential planning data before starting implementation.

Auto-compaction triggers arbitrarily based on token limits rather than task phases, dropping important context mid-workflow. Strategic compaction prompts you at task boundaries, ensuring completed research and plans are safely preserved before moving to the next phase.

You manage token limits in multi-phase tasks by triggering manual context compaction at phase boundaries. This approach tracks tool calls and prompts you to compact strategically, freeing up context space before starting resource-intensive implementation phases.

Yes, you need a shell environment to execute context management scripts and configure environment variables. These scripts track tool calls and monitor configurable thresholds to prompt manual compaction at the right moments.

You should manually compact the AI context window at logical task intervals, such as when switching from research to planning or before coding begins. This prevents arbitrary auto-compaction from discarding essential context mid-phase.