strategic-compact

Prevent AI context loss during long coding sessions by using phase-aware compaction prompts that suggest manual context compaction at logical task boundaries instead of arbitrary auto-trigger points, retaining critical plans while discarding exploratory data.

Context compaction is the process of summarizing or discarding old context to free up context window space. You should use it in multi-phase tasks at logical boundaries, such as after finalizing a plan and before starting implementation, to retain essential context.

You can configure context window optimization by setting a threshold for Edit and Write tool calls. The system monitors these calls and provides periodic reminders to compact context after your configured threshold is reached during workflows.

Yes, arbitrary auto-compaction interrupts multi-phase AI workflows by triggering at random points, which risks losing critical context. Phase boundary compaction guidance prevents this by prompting compaction only at logical transitions between unrelated work items.

The best way to manage context shifts between unrelated work items is applying phase boundary compaction guidance. This approach prompts you to compact context specifically when transitioning between tasks, freeing up context window space safely.

You should avoid automatic context compaction in AI pipelines when transitioning between critical phases like research and implementation. Arbitrary auto-compaction risks losing finalized plans, so manual phase-aware compaction is recommended instead.