Claude Code v2.1.277 Adds AGENTS.md Support and Fixes Crash Bugs
Claude Code v2.1.277 landed with a small set of feature additions and a long list of bug fixes across sessions, tools, and gateway handling. Here's what actually changed.
AGENTS.md support
In a project with no CLAUDE.md, Claude Code now reads AGENTS.md instead. You can switch it under "Project instructions" in /config. Not yet available on Bedrock, Vertex, or Foundry.
Gateway and proxy changes
CLAUDE_GATEWAY_PROXY_IS_EGRESS_BOUNDARY=1for Claude apps gateways whose only egress is a forward proxy: every outbound request hands the proxy the hostname instead of resolving it locally.- Optional
headers:map on Claude apps gateway upstreams to send static headers to a proxy you run in front of a provider. - A line now notes that a background task's update is waiting when it finishes while a panel such as
/tasksis open.
Session and CLI fixes
claude -pand Agent SDK sessions that could hang with no result after an internal error now report the error and exit with code 1.- Conversations failing every request with "text content blocks must be non-empty" when an earlier assistant turn held an empty text block beside other content, including after
--resume. - Unexpected logout when an older Claude Code build (e.g., an IDE extension's bundled CLI) runs on the same machine as the current one.
- Interactive start-up hanging or erroring for
ANTHROPIC_API_KEYusers when~/.claude.jsonholds a malformedcustomApiKeyResponsesvalue. - Update checks erroring every 30 minutes, and
claude updatehanging when a min/max version is set, if a proxy returns an invalid version. A malformedminimumVersionis now ignored. claude updateon winget- or apk-managed installs reporting "up to date" when the version lookup failed.claude plugin installsometimes failing and breaking the installed copy when reinstalling a plugin version in use; an unchanged copy is now left alone.
Tool fixes (Grep, Glob, Write, Edit)
- Grep and Glob now return an error when the search could not start due to out of processes, memory, or file handles, instead of reporting no matches.
- Write no longer silently ends the turn as a declined permission when the target path is an existing directory.
- Edit no longer treats an escaped backslash followed by
uXXXXas a\uXXXXescape, which could rewrite an escaped backslash sequence when editing non-ASCII characters. - Edit reports "String not found in file" instead of "Invalid regular expression: regular expression too large" for very large non-ASCII edits that don't match.
- Turns no longer end early with "Path contains null bytes" when a tool call's file path contains
\u0000written as an escape sequence.
Stability fixes
- Background sessions (
claude --bg) no longer exit when a plugin's LSP server exits or closes stdin. - Crashes from malformed
claudeAiMcpEverConnected,theme, or a plain-string assistant message in saved history. - Crash ("unrecoverable interface error") when the prompt contained terminal color codes.
- Sessions on slow machines exiting when the first spinner appeared.
- Rare screen freeze after an internal rendering error.
- Windows turns stopping with "Out of memory" right after a reply, leaving tool calls unrun.
- Sessions continued after
/clear(restart,--continue,--resume) missing part of the first message when a SessionStart hook printed output, causing a full prompt-cache miss.
If you run Claude Code behind a proxy or use plugins with LSP servers, this release is worth pulling.
📖 Read the full source: GitHub Claude-Code
👀 See Also

STAR Reasoning Framework Accuracy Drops from 100% to 0% in Production Prompts
A researcher found that the STAR reasoning framework, which raised Claude's accuracy on an implicit constraint problem from 0% to 100% in isolation, dropped to 0-30% accuracy when used inside a 60-line production system prompt. The issue was caused by conflicting instructions in the production prompt that triggered premature answer commitments.

Qwen3.6 Plus benchmark comparison against Western SOTA models
Qwen3.6 Plus scores 78.8 on SWE-bench Verified, 90.4 on GPQA/GPQA Diamond, 28.8 on HLE (no tools), and 78.8 on MMMU-Pro, placing it competitively against models like GPT-5.4, Claude Opus 4.6, and Gemini 3.1 Pro Preview.

Polaroid's Anti-AI Campaign Targets Digital Fatigue with Analog Print Ads
Polaroid's new Flip camera campaign uses print ads with copy like 'AI can't generate sand between your toes' to contrast analog with digital/AI culture.

Claude Status Update: Elevated Error Rates for Opus 4.6 and Sonnet 4.6
An official Claude system status update reports elevated error rates for Opus 4.6 and Sonnet 4.6 models, with an incident timestamped 2026-03-31T21:10:28.000Z. The automatic post directs users to check resolution status and community performance reports.