strategic-compact

Manual context compaction prevents AI context loss by suggesting targeted `/compact` commands at logical task intervals. This approach preserves critical information by prompting you to clear context strategically between phases, rather than relying on arbitrary auto-compaction that might drop key details.

Manual context compaction is the process of intentionally clearing AI memory at logical workflow breakpoints. Unlike arbitrary auto-compaction, it tracks tool calls and prompts you to compact context only when a configurable threshold is reached, ensuring task phase continuity.

Set up automated context management by configuring a shell script in your `~/.claude/settings.json` hooks. Attach the script to the `PreToolUse` event for `Edit` or `Write` tools so it tracks usage and suggests compaction when thresholds are met.

Yes, you can configure the token usage threshold for context compaction prompts. The system tracks tool calls against this configurable limit and provides periodic reminders to compact context, aligning memory management with your specific task phases.

You should manually compact context at logical workflow transitions, such as after completing research and before writing a report. This prevents arbitrary auto-compaction from discarding important implementation plans or research notes mid-task.

Yes, this context management approach works without external dependencies. It relies solely on a configured shell script integrated into your existing settings hooks to monitor tool calls and trigger manual compaction suggestions.