Context Death Spiral Prevention — OpenClaw Compaction Primer

What Is a Context Death Spiral?

A context death spiral is what happens when an OpenClaw agent accumulates so much conversation history that its reasoning quality degrades — and then the degradation makes it handle the accumulation worse, which accelerates the degradation.

You've seen the symptoms:

• Agent starts forgetting instructions it acknowledged 20 turns ago
• Response quality drops noticeably mid-session without any obvious trigger
• Agent begins contradicting itself or repeating earlier failed attempts
• Sudden unexplained context resets that wipe work in progress
• Tool calls become erratic — the agent loses track of what it already tried

These aren't model failures. They're architecture failures. The agent isn't broken — its context management is.

Why Default OpenClaw Setups Don't Handle This