Feature Request: Native context persistence across compaction events

Problem

The new background task/subagent capabilities are powerful but exhaust context windows rapidly. When running multiple parallel agents or long research tasks, we frequently hit compaction events that lose critical session context - causing:

• Session failures: Agents lose track of what they were doing mid-task
• Repeated work: Claude re-discovers the same information after compaction
• Broken workflows: Multi-step tasks fail when context compacts between steps

This is especially problematic with the Task tool spawning subagents that each consume significant context.

Impact

We're finding that 30+ minute sessions with multiple subagents almost always hit compaction, losing accumulated context. Complex multi-agent workflows become unreliable.

Current Workaround

We built a minimal open-source memory bridge using the hooks API:

https://github.com/Equilateral-AI/claude-memory-bridge

It uses PreCompact and SessionStart hooks to save/restore basic session context. It helps, but it's limited by what the hooks API exposes.

Request

Some form of native context persistence across compaction events would significantly improve reliability for power users running complex multi-agent workflows.

Happy to provide more details or test proposed solutions.