DevAgentRadar → Claude Code

You are helping me adopt a real coding-assistant change. Work only from the facts below. Do not invent features.

Context

Assistant: Claude Code Proposal: Upgrade Claude Code to v2.1.273 to restore full-window auto-compact Summary: v2.1.273 stops auto-compact firing at ~half the window, adds opt-in LLM gateway hint headers, MCP disconnect notices that point to /mcp, remote-control session forking, working MDM MCP denylists, and richer OTEL names. Upgrade, then set CLAUDE_CODE_GATEWAY_HINT_HEADERS=1. Primary source: https://github.com/anthropics/claude-code/releases/tag/v2.1.273

Why it matters

v2.1.273 fixed the context meter and auto-compact counting advisor-tool turns at roughly twice their real size, which made auto-compact fire at about half the real window. Upgrade so long sessions keep working until the actual window is full. If you front Claude Code with an LLM gateway, set CLAUDE_CODE_GATEWAY_HINT_HEADERS=1 to send x-claude-code-request-class, x-claude-code-agent-type, x-claude-code-prev-tool-durations, x-claude-code-compaction, and x-claude-code-context-compacted. MCP servers that drop mid-session now notify you when automatic reconnection gives up and point at /mcp. A session started with claude --remote-control or /remote-control can be forked from the Claude app as a background session on your computer. allowManagedMcpServersOnly, deniedMcpServers, and disableClaudeAiConnectors from MDM or managed-settings.json were ignored when server-managed settings were also present; they now apply, so re-deploy those files. OTEL_LOG_TOOL_DETAILS=1 now includes real agent, skill, plugin, and MCP server names on cost and token metrics. Auto mode on Bedrock, Vertex, and Foundry now uses the local classifier by default; set CLAUDE_CODE_AUTO_MODE_SERVER=1 only to use the platform server-side classifier. Ignore the +28 omitted changelog entries.

Suggested actions

1. Upgrade Claude Code to v2.1.273 and restart so auto-compact and the context meter stop counting advisor-tool turns at ~2x (they previously fired compact at about half the real window).
2. Add CLAUDE_CODE_GATEWAY_HINT_HEADERS=1 to the Claude Code environment and restart so LLM gateways receive x-claude-code-request-class, x-claude-code-agent-type, x-claude-code-prev-tool-durations, x-claude-code-compaction, and x-claude-code-context-compacted.
3. Set OTEL_LOG_TOOL_DETAILS=1 so cost and token metrics include the real agent, skill, plugin, and MCP server names.
4. Re-deploy MDM or managed-settings.json values for allowManagedMcpServersOnly, deniedMcpServers, and disableClaudeAiConnectors, then restart Claude Code — v2.1.273 applies them even when server-managed settings are also present.
5. When Claude Code reports an MCP disconnect and auto-reconnect has given up, run /mcp in that session to re-authenticate. From a session started with claude --remote-control or /remote-control, fork in the Claude app so the fork runs as a background session on this computer. On Bedrock, Vertex, or Foundry, leave the new local auto-mode classifier, or set CLAUDE_CODE_AUTO_MODE_SERVER=1 only if you need the platform server-side classifier.

Config surfaces this release may change

• MCP servers (high confidence) — check your repo before applying
• permission rules (high confidence) — check your repo before applying
• settings files (high confidence) — check your repo before applying
• skills (high confidence) — check your repo before applying
• context and compaction settings — check your repo before applying
• subagent definitions — check your repo before applying

After you finish

Do not report this as applied to DevAgentRadar. You cannot write the visitor's loop.

Tell the human: open https://devagentradar.com/proposals/claude-code-v2-1-273-upgrade-claude-code-to-v2-1-273-to-restore-full-win and mark Applied, Skipped, or Failed. Proposal id: e2b0c496-61bc-4acb-b841-574b22ff8c47

Your job

1. Restate the change in one sentence.
2. Propose a minimal plan for my repo (or a throwaway pilot).
3. Implement only what I approve; prefer small diffs and tests.
4. Call out risks (permissions, breaking APIs, cost).

Start by confirming you understood the proposal.