The Claude Code Changelog

Эпизоды

• 
  Claude Code 2.1.277: Safer Subagents and Better CLI Diagnostics 20.09.2026

  Claude Code 2.1.277 tackles prompt injection in multi-agent workflows with structural subagent result framing, making it harder for untrusted text to hijack orchestration. The episode also covers better headless error handling, invisible prompt sanitization, and clearer Grep/Glob diagnostics.
• 
  Claude Code’s Cheaper Auto Mode and the Proxy Trap 19.09.2026

  Anthropic’s latest Claude Code update moves safety classification checks to the server, cutting token costs for auto mode unless a proxy strips the required headers and forces a fallback. The episode also covers how to diagnose eligibility with /status, disable server checks with an environment variable, and the new AGENTS.md fallback for repo instructions. Show Notes Claude Code 2.1.275 — claude.ai skills and plugins sync to the terminal: https://ai-tldr.dev/releases/anthropic-claude-code-2-1-275/ Auto mode classifier request charges - Claude Code Docs: https://code.claude.com/docs/en/auto-mode-classifier-billing
• 
  Claude Code Gets a Send Now Key 18.09.2026

  Anthropic’s latest Claude Code updates make agent steering much smoother, adding a send now shortcut that interrupts a running turn without losing context. The episode also covers account-wide skill and plugin syncing, marketplace install improvements, and a quick hotfix for enterprise gateway and proxy users.
• 
  Claude Code 2.1.274 Fixes Death Loops and Goal Loss 17.09.2026

  We break down Claude Code 2.1.274’s biggest workflow fixes, including automatic repair for desynced tool calls, cleaner failure handling with slash rewind, and better goal persistence through context compaction. Plus, a look at the new default Model Context Protocol client on Bedrock, Vertex, and Foundry, and how to keep legacy MCP setups running if you need them.
• 
  Claude Code 2.1.273: Gateway Hints for Smarter Routing 16.09.2026

  We break down Anthropic’s latest Claude Code update, including opt-in gateway hint headers that let proxies route traffic without inspecting private prompt bodies. Plus, we cover cleaner MCP reconnect notifications, remote session forking, and the return of local default classification for Bedrock and Vertex auto mode.
• 
  Claude Code 2.1.271: Leaner Subagents and Tighter Sandboxes 15.09.2026

  We break down Anthropic’s Claude Code 2.1.271 update, including the new omitClaudeMd setting that keeps subagents focused by skipping local context files while preserving managed policies. The episode also covers tighter command-level sandboxing, SHA-256 plugin verification, and nicer-looking Markdown artifacts.
• 
  Claude Code’s WebFetch Timeout Fix 14.09.2026

  We break down the Claude Code 2.1.268 fix for WebFetch hangs, including the new 300-second deadline and the CLAUDE_CODE_WEBFETCH_DEADLINE_MS setting for tuning timeouts in CI. The episode also covers session cleanup for self-hosted runners, JSON output for plugin commands, and tighter local file safety in Cowork sessions.
• 
  Claude Code 2.1.270: Inline Diffs and Key Fixes 13.09.2026

  We break down how Claude Code 2.1.270’s bashEditDiffEnabled surfaces inline diffs for shell edits, helping the agent see file changes instantly without extra git diff calls. The episode also covers a tee-related security fix, a read-only git permission regression, and improved retry handling for more reliable terminal workflows.
• 
  Claude Code 2.1.269: Plugin Evals and 256-Agent Workflows 12.09.2026

  Claude Code 2.1.269 brings a dedicated plugin eval CLI for reproducible benchmarking, with scored JSON output and an HTML report to catch regressions before shipping. The episode also covers higher workflow concurrency via CLAUDE_CODE_WORKFLOW_MAX_CONCURRENT_AGENTS and new /output style controls for cleaner remote and headless runs.
• 
  Claude Code SessionEnd Hooks Fixed: No More 1.5s Teardown Cuts 11.09.2026

  We break down a Claude Code bug that caused SessionEnd hooks to be killed after just 1.5 seconds, even when a global timeout was set, leading to truncated teardowns, missing uploads, and corrupted logs. The episode also covers the fix, best practices for per-hook timeouts, and related updates for gateway pricing sync and OpenTelemetry exports.
• 
  Claude Code 2.1.267: Hard-Cap AI Thinking Costs 10.09.2026

  We break down how Claude Code 2.1.267 helps teams control AI spending with a hard cap on extended thinking across cloud providers, while still letting subagents and custom prompts use lower effort levels when appropriate. The episode also covers the new system prompt snapshot toggle, giving prompt engineers faster live iteration without restarting sessions.
• 
  Claude Code Adds Hot-Reloaded Local Plugin Bundles 09.09.2026

  Claude Code 2.1.265 brings hot-reloading for local plugins, letting you point plugin dir at a parent folder and load or unload multiple tools live without restarting your session. The episode also covers symlink safety checks, clearer truncation notices for large tool logs, and a refreshed workflows UI for debugging nested agent activity.
• 
  Locking Down AI Auto Mode and Read Access 08.09.2026

  We break down new safeguards for AI coding agents in auto mode, including out-of-bounds read protection, stricter cloud metadata containment, and safer behavior in headless CI runs. The episode also covers terminal usability upgrades and clearer diagnostics for proxy and policy issues.
• 
  Claude Code’s Inline Output Buffers Slash Tool Round Trips 07.09.2026

  Claude Code gets a major workflow upgrade with larger inline output buffers that keep long test failures and stack traces in context, cutting out extra file-read tool calls. The episode also covers the token tradeoffs, smarter test filtering, and a handful of CLI polish improvements like subagent prompt files, Bash-style keybindings, and stronger safety checks. Show Notes https://raw.githubusercontent.com/anthropics/claude-code/refs/heads/main/CHANGELOG.md: https://raw.githubusercontent.com/anthropics/claude-code/refs/heads/main/CHANGELOG.md
• 
  Claude Code 2.1.260: Uncapped Subagents and Safer Worktrees 06.09.2026

  This episode covers Claude Code 2.1.260’s biggest workflow upgrade: removing the one-hour limit on subagent background tasks, plus new ways to monitor long-running jobs with attach, logs, and stop. It also highlights the new text-based advisor command and safer git worktree deletion prompts that help prevent accidental data loss. Show Notes Claude Code changelog - Claude Code Docs: https://code.claude.com/docs/en/changelog
• 
  Claude Code Skill Doctor Cuts Hidden Token Bloat 05.09.2026

  Claude Code 2.1.261 adds skill doctor to reveal unused loaded skills and their context costs, helping you prune token-heavy workspace setups. The episode also covers larger inline output limits, subagent prompt loading from files, and stronger rm safety checks. Show Notes https://raw.githubusercontent.com/anthropics/claude-code/refs/heads/main/CHANGELOG.md: https://raw.githubusercontent.com/anthropics/claude-code/refs/heads/main/CHANGELOG.md
• 
  Claude Code Gets Live Diff Views and Smarter Cache Debugging 04.09.2026

  Anthropic’s latest Claude Code release brings a live split-screen diff panel right into the terminal, making it easier to watch code changes as they happen. The hosts also dig into improved prompt cache miss diagnostics, headless session upgrades, and a security fix for auto-approval in zsh scripts.
• 
  Claude Code 2.1.259 Fixes CI Prompt Hangups 03.09.2026

  We break down Claude Code 2.1.259’s new unattended-host workflow, including --permission-prompts none, how denied prompts keep CI jobs moving, and what to watch for with preapproved tools. The episode also covers the parallel-session state fix, plus quality-of-life updates for GitLab MR commands and plugin validation.
• 
  Hard-Locking Claude Subagents to Stop Surprise Bills 02.09.2026

  This episode breaks down new Claude Code workflow updates, including how subagent model precedence now works and why the new FORCE environment variable can hard-lock every subagent to a chosen model. It also covers session-level effort controls and new containment rules that add stronger cost and security guardrails for autonomous agents.
• 
  Claude Code Gets Unix-Style Background Session Commands 01.09.2026

  This episode breaks down the new direct CLI controls for Claude Code background sessions, including attach, logs, stop, respawn, and rm, plus how they streamline scripting and terminal workflows. It also covers key fixes in versions 2.1.251 and 2.1.252, from git worktree support and macOS task handling to permission saving and cleaner large-error notifications.