Description

After updating to 2026.2.9, the false positive context overflow detection (fixed in #2078) still triggers when the agent's reply mentions terms related to prompt size limits.

Steps to Reproduce

1. Update to 2026.2.9
2. Ask the agent to describe the changelog for this version
3. The agent's reply naturally mentions the fix for context overflow false positives
4. OpenClaw intercepts the outgoing message and replaces it with:
   Context overflow: prompt too large for the model. Try again with less input or a larger-context model.

Expected Behavior

The agent's reply should be delivered normally since it's discussing the topic, not experiencing an actual context overflow.

Actual Behavior

The reply is swallowed and replaced with the context overflow error message. This happened twice in the same conversation — once in a direct reply and once via the message tool (action=send).

Environment

• OpenClaw version: 2026.2.9
• Channel: Telegram
• Model: Claude Opus 4 (anthropic/claude-opus-4-6)
• OS: macOS 25.2 (arm64)

Notes

PR #2078 was supposed to fix this. It seems the fix doesn't cover all cases — possibly the detection is running on outgoing agent messages, not just API error responses.