docs: Runtime stability issues observed in production (macOS + EvoMap Hub)

[PangTianHua]

Summary

Production run on macOS Darwin 25.3.0 (arm64), Node v24.11.1, evolver v1.41.0, OpenClaw 2026.4.2.

Documented 7 issues ranging from Critical to Low severity, with root cause analysis and suggested fixes.

Issues

#	Issue	Severity
1	System load backoff loop (LOAD_MAX default too low for macOS)	Critical
2	node_secret_invalid → rate limit death spiral	Critical
3	EVOLVE_BRIDGE=false silent rejection (appears to work but doesn't)	High
4	ClawHub nativeSkills:auto destroys local symlink install	Critical
5	LLM fallback to unsupported model variant	Medium
6	Duplicate heartbeat cron conflict	Low
7	Hub API timeout instability	Low

Details

See docs/runtime-stability-issues.md for full root cause analysis and suggested fixes for each issue.

[autogame-17]

Thanks @PangTianHua! This is an excellent production stability report. Several of the issues you documented have already been partially addressed:

• Issue 1 (LOAD_MAX): v1.44.0 supports EVOLVE_LOAD_MAX env var to override the default threshold. We should add per-platform defaults as you suggest.
• Issue 2 (rate limit spiral): v1.43.0 added heartbeat concurrency guards. The exponential backoff suggestion for 403s is good -- we'll add that.
• Issue 3 (EVOLVE_BRIDGE=false): Valid concern. Will add a startup warning when bridge=false in loop mode.

Merged into main with co-author attribution (commit 2affcbd). Closing this PR as the doc has been applied via our internal workflow. Your feedback is very valuable -- thank you!