fix(server): rotate Claude on hard usage rejection

[bk-agent-01]

Closes TEC-961

What Changed

• Listen for authoritative Claude SDK rate_limit_event rejections with a supported hard-limit type.
• Invoke the host claude-autoswitch --hard-limit <type> --json contract and recycle only the affected provider thread after a validated switch.
• Debounce repeated evidence, preserve unrelated sessions, and leave sessions running on no-op or failure paths.
• Document the 30-second fallback timer, immediate rejection flow, staging checks, and rollback.

Why

Claude can reject a request before its cached utilization reflects exhaustion. The fallback timer therefore cannot always rotate accounts promptly. This hook uses the typed provider rejection as authoritative evidence while keeping credential selection in the host-owned elector.

Staging Evidence

• PR fix(server): rotate Claude on hard usage rejection #125 CI passed, including focused handler tests: 7/7.
• Tests inject the real provider event and prove one election call, one affected-thread recycle, debounce, unrelated-session preservation, and safe no-op/failure handling.
• Before the selector proof, default was elected at 34% five-hour usage, below the 85% trip threshold.
• /home/ubuntu/.local/bin/claude-autoswitch --hard-limit five_hour --json exited 0 with status=switched, from=default, and to=agent.
• Autoswitch state recorded last_hard_limit_type=five_hour and last_to=agent; the timer was restored active and enabled.
• A fresh ExpBkT3 Claude probe started after election and replied exactly RECOVERED in 5.265 seconds with lastError=null, then stopped cleanly.
• expbkmain deployment SHA is a12718bda83ddd28ab2fdcf4bd6dda59b1bb0b2d; staging service and HTTP checks passed.

Checklist

• This PR is small and focused
• I explained what changed and why
• UI screenshots are not applicable
• Interaction video is not applicable

Generated with GPT-5.6-SOL in T3 Code.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

ℹ️ No successful main baseline artifact is available yet. This run establishes the initial measurement.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	—	11.4 KiB	—	15.1 KiB	✅
Codex	Thread snapshot wire	—	5.7 KiB	—	7.3 KiB	✅
Codex	Live turn WebSocket wire	—	5.7 KiB	—	7.8 KiB	✅
Codex	Live turn WebSocket decoded	—	50.3 KiB	—	66.4 KiB	✅
Codex	Live turn messages	—	9	—	21	✅
Claude	Total thread wire	—	11.4 KiB	—	15.1 KiB	✅
Claude	Thread snapshot wire	—	5.7 KiB	—	7.3 KiB	✅
Claude	Live turn WebSocket wire	—	5.7 KiB	—	7.8 KiB	✅
Claude	Live turn WebSocket decoded	—	51.2 KiB	—	66.4 KiB	✅
Claude	Live turn messages	—	10	—	21	✅

Baseline: unavailable · PR result: a12718b · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 95.6 KiB
• Claude decoded thread snapshot: 96.3 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.