Claude Code v2.1.283 introduces new configuration options for LLM gateways and model access control, alongside a tool to audit prompting patterns in CLAUDE.md files.
- Added x-claude-code-prompt-id header to group requests by user prompt via CLAUDE_CODE_GATEWAY_HINT_HEADERS=1.
- Introduced availableModelsMatch managed setting with "exact" mode to block new model versions until explicitly listed.
- Added deniedModels managed setting to block specific models even if allowed by availableModels.
- Added /doctor prompt-audit and /checkup prompt-audit commands to detect prompting patterns written for older models.
- Fixed SDK sessions losing deferred tool calls or results when turns end early or after worker restarts.
- Improved startup latency by deferring auto-mode classifier rules and Artifact tool loading to first use.
These changes provide better control over model access, gateway request grouping, and configuration hygiene while fixing critical session stability issues.