Claude Code v2.1.283 (2026-09-25) adds availableModelsMatch=exact and deniedModels for managed model allow/deny lists, opt-in x-claude-code-prompt-id gateway hint headers, /doctor prompt-audit for CLAUDE.md/skills/agents/commands, plus Bedrock Mantle upstream and load_test_mode on the Apps Gateway. npm @anthropic-ai/claude-code 2.1.283 is live.

Key Takeaways

  • ✓Shipped: GitHub v2.1.283 = npm @anthropic-ai/[email protected]
  • ✓Governance: availableModelsMatch=exact pins exact model versions; deniedModels blocks models even if allowlisted
  • ✓Observability: CLAUDE_CODE_GATEWAY_HINT_HEADERS=1 adds x-claude-code-prompt-id; OTEL_LOG_TOOL_CONTENT=1 logs tool outputs on spans
  • ✓Audit: /doctor prompt-audit (alias /checkup) reviews CLAUDE.md, skills, agents, commands for older-model patterns
  • ✓Gateway: Bedrock Mantle upstream + opt-in load_test_mode for signed canned replies
🔗

Project Links & Resources

Direct Access
Direct access to official project resources and documentation
🐙
Claude Code v2.1.283 Release
github.com/anthropics/claude-code/releases/tag/v2.1.283
↗
📖
@anthropic-ai/claude-code 2.1.283 on npm
www.npmjs.com/package/@anthropic-ai/claude-code/v/2.1.283
↗
🐙
Claude Code GitHub repo
github.com/anthropics/claude-code
↗
🐙
Prior v2.1.282 Release
github.com/anthropics/claude-code/releases/tag/v2.1.282
↗
🔬

In-Depth Technical Analysis

Background Enterprise coding agents struggle with model-allowlist drift, gateways that cannot group requests by one user prompt, CLAUDE.md/skills written for older models, and fragile MCP progress/session recovery. After 2.1.282's prose/Chrome/resume fixes, Claude Code v2.1.283 (2026-09-25) pushes governance, observability, and gateway load testing forward. ### Architecture Managed setting availableModelsMatch=exact restricts availableModels entries to the exact version named; deniedModels can still block allowlisted models. Opt-in CLAUDE_CODE_GATEWAY_HINT_HEADERS=1 injects x-claude-code-prompt-id for gateway aggregation. With OTEL_LOG_TOOL_CONTENT=1, MCP/WebFetch/WebSearch outputs land on tool.output span events. /doctor prompt-audit (alias /checkup) reviews CLAUDE.md, skills, agents, and commands for older-model prompting patterns. Apps Gateway adds Bedrock Mantle upstream and load_test_mode (sign but do not send; canned replies). Fixes cover deferred tool loss on early turn end, discarded background MCP progress, hung stdio MCP on session end, and one-shot 404 leaving remote MCP falsely connected. ### Benchmarks No official SWE-bench numbers. Measure model-resolve failures under exact allowlists and gateway aggregation completeness with prompt-id headers. ### Getting started npm i -g @anthropic-ai/[email protected]. Configure availableModelsMatch/deniedModels; set CLAUDE_CODE_GATEWAY_HINT_HEADERS=1; run /doctor prompt-audit. Full notes on the GitHub release.