Anthropic 发布了其开发工具 Claude Code 的多个更新，涵盖版本 v2.1.141 至 v2.1.150。这些更新为后台会话管理、插件功能和工具集成带来了显著改进，尤其针对 Windows 用户。主要增强功能包括更好地处理空闲会话、更可靠的自动更新程序错误报告，以及用于配置后台代理的扩展命令行选项。此次发布还解决了与权限、沙盒和用户界面响应能力相关的多项错误，旨在提供更稳定、更高效的编码环境。
AI
<h2>What's changed</h2> <ul> <li>Introducing Claude Fable 5: a Mythos-class model that we’ve made safe for general use. Fable’s capabilities exceed those of any model we’ve ever made generally available. Update to version 2.1.170 for access. <a href="https://www.anthropic.com/new…
<h2>0.108.0 (2026-06-09)</h2> <p>Full Changelog: <a href="https://github.com/anthropics/anthropic-sdk-python/compare/v0.107.1...v0.108.0">v0.107.1...v0.108.0</a></p> <h3>Features</h3> <ul> <li><strong>api:</strong> add support for claude-mythos-5 and claude-fable-5, with support …
<h2>What's changed</h2> <ul> <li>Added <code>--safe-mode</code> flag (and <code>CLAUDE_CODE_SAFE_MODE</code>) to start Claude Code with all customizations (CLAUDE.md, plugins, skills, hooks, MCP servers) disabled for troubleshooting</li> <li>Added <code>/cd</code> command to move…
<p>Ollama Launch now supports Hermes Desktop, a native desktop interface for the Hermes agent. Run it alongside your Hermes agent to get a visual interface for managing conversations, integrations, and messaging apps.</p> <div class="snippet-clipboard-content notranslate position…
<h2>What's changed</h2> <ul> <li>Added <code>fallbackModel</code> setting to configure up to three fallback models tried in order when the primary model is overloaded or unavailable; <code>--fallback-model</code> now also applies to interactive sessions</li> <li>Added glob patter…
<h2>0.106.0 (2026-06-05)</h2> <p>Full Changelog: <a href="https://github.com/anthropics/anthropic-sdk-python/compare/v0.105.2...v0.106.0">v0.105.2...v0.106.0</a></p> <h3>Features</h3> <ul> <li><strong>api:</strong> mark Claude Opus 4.1 as deprecated (<a href="https://github.com/a…
<h2>What's changed</h2> <ul> <li>Added <code>requiredMinimumVersion</code> and <code>requiredMaximumVersion</code> managed settings — Claude Code refuses to start if its version is outside the allowed range and directs the user to an approved version</li> <li>Added <code>/plugin …
<h2>What's changed</h2> <ul> <li><code>claude agents --json</code> now includes <code>waitingFor</code> showing what a waiting session is blocked on (e.g. permission prompt)</li> <li><code>--tools</code>: explicitly listing Grep/Glob now provides the dedicated search tools on nat…
<h2>What's Changed</h2> <ul> <li>models: add support for gemma4-12b by <a class="user-mention notranslate" href="https://github.com/pdevine">@pdevine</a> in <a class="issue-link js-issue-link" href="https://github.com/ollama/ollama/pull/16457">#16457</a></li> </ul> <p><strong>Ful…
<h2>What's changed</h2> <ul> <li><code>OTEL_RESOURCE_ATTRIBUTES</code> values are now included as labels on metric datapoints, so you can slice usage metrics by custom dimensions like team or repo</li> <li><code>claude agents</code> rows now show <code>done/total</code> before th…
<h2>What's changed</h2> <ul> <li>Added a prompt before writing to shell startup files (<code>.zshenv</code>, <code>.zlogin</code>, <code>.bash_login</code>) and <code>~/.config/git/</code>, which could otherwise lead to unintended command execution</li> <li><code>acceptEdits</cod…
<h2>What's changed</h2> <ul> <li>Auto mode is now available on Bedrock, Vertex, and Foundry for Opus 4.7 and Opus 4.8. Opt in by setting <code>CLAUDE_CODE_ENABLE_AUTO_MODE=1</code></li> </ul>
<h2>What's changed</h2> <ul> <li>Plugins in <code>.claude/skills</code> directories are now automatically loaded, no marketplace required</li> <li>Added <code>claude plugin init <name></code> to scaffold a new plugin in <code>.claude/skills</code></li> <li>Added autocomplet…
<h2>What's changed</h2> <ul> <li>Opus 4.8 is here! Now defaults to high effort · /effort xhigh for your hardest tasks</li> <li>Introducing dynamic workflows: ask Claude to create a workflow and it orchestrates work across tens to hundreds of agents in the background, so you can t…
<h2>What's changed</h2> <ul> <li>Added <code>skipLfs</code> option to <code>github</code>/<code>git</code> plugin marketplace sources to skip Git LFS downloads during clone and update</li> <li>Claude Code now shows a one-time notice when your npm global install can't auto-update;…
<p>This version of Ollama will change the architecture to directly support llama.cpp instead of building on top of GGML, and allows for compatibility with GGUF file format. MLX is used to accelerate model inference on Apple Silicon.</p> <p>While in pre-release we'd love <a href="…
<p>This version of Ollama will change the architecture to directly support llama.cpp instead of building on top of GGML, and allows for compatibility with GGUF file format. MLX is used to accelerate model inference on Apple Silicon.</p> <p>While in pre-release we'd love <a href="…
<h2>What's changed</h2> <ul> <li><code>/code-review --fix</code> now applies review findings to your working tree after the review, surfacing reuse, simplification, and efficiency suggestions; <code>/simplify</code> now invokes <code>/code-review --fix</code></li> <li>Skills and …
<h2>What's changed</h2> <ul> <li><code>/usage</code> now shows a per-category breakdown of what's driving your limits usage — skills, subagents, plugins, and per-MCP-server cost</li> <li><code>/diff</code> detail view can now be scrolled with the keyboard (arrows, <code>j</code>/…
<p>This version of Ollama will change the architecture to directly support llama.cpp instead of building on top of GGML, and allows for compatibility with GGUF file format. MLX is used to accelerate model inference on Apple Silicon.</p> <p>While in pre-release we'd love <a href="…
<h2>What's changed</h2> <ul> <li>Fixed the Bash tool returning exit code 127 on every command for some users (a regression introduced in 2.1.147)</li> </ul>
<h2>What's changed</h2> <ul> <li>Added the <code>Workflow</code> tool for deterministic multi-agent orchestration. It is off by default — set <code>CLAUDE_CODE_WORKFLOWS=1</code> to enable</li> <li>Pinned background sessions (<code>Ctrl+T</code> in <code>claude agents</code>) now…
<h2>What's changed</h2> <ul> <li>Renamed <code>/simplify</code> to <code>/code-review</code> with an optional effort level (e.g. <code>/code-review high</code>)</li> <li>Auto mode no longer suppresses <code>AskUserQuestion</code> when the user or a skill explicitly relies on it</…
<h2>What's changed</h2> <ul> <li>Added <code>claude agents --json</code> to list live Claude sessions as JSON for scripting (tmux-resurrect, status bars, session pickers)</li> <li>Added <code>agent_id</code> and <code>parent_agent_id</code> attributes to <code>claude_code.tool</c…
<h2>What's changed</h2> <ul> <li>Added <code>/resume</code> support for background sessions — sessions started via <code>claude --bg</code> or agent view now appear alongside interactive ones, marked with <code>bg</code></li> <li>Added elapsed duration to background subagent comp…
<h2>What's changed</h2> <ul> <li>Added plugin dependency enforcement: <code>claude plugin disable</code> now refuses when another enabled plugin depends on the target (with a copy-pasteable disable-chain hint), and <code>claude plugin enable</code> force-enables transitive depend…
<h2>What's changed</h2> <ul> <li>Added <code>terminalSequence</code> field to hook JSON output so hooks can emit desktop notifications, window titles, and bells without a controlling terminal</li> <li>Added <code>CLAUDE_CODE_PLUGIN_PREFER_HTTPS</code> to clone GitHub plugin sourc…
<h2>What's changed</h2> <ul> <li>Added agent view (Research Preview): a single list of every Claude Code session — running, blocked on you, or done. Run <code>claude agents</code> to get started. See <a href="https://code.claude.com/docs/en/agent-view" rel="nofollow">https://code…
<h2>What's changed</h2> <ul> <li>Added <code>CLAUDE_CODE_ENABLE_FEEDBACK_SURVEY_FOR_OTEL</code> to re-enable the session quality survey for enterprises capturing responses through OpenTelemetry</li> <li>Added <code>settings.autoMode.hard_deny</code> for auto mode classifier rules…
<h2>What's changed</h2> <ul> <li>Added <code>worktree.baseRef</code> setting (<code>fresh</code> | <code>head</code>) to choose whether <code>--worktree</code>, <code>EnterWorktree</code>, and agent-isolation worktrees branch from <code>origin/<default></code> or local <cod…
<h2>What's changed</h2> <ul> <li>Added <code>CLAUDE_CODE_SESSION_ID</code> environment variable to the Bash tool subprocess environment, matching the <code>session_id</code> passed to hooks</li> <li>Added <code>CLAUDE_CODE_DISABLE_ALTERNATE_SCREEN=1</code> env var to opt out of t…
<h2>What's changed</h2> <ul> <li>Fixed VS Code extension failing to activate on Windows due to a hardcoded build path in the bundled SDK (<code>createRequire</code> polyfill bug)</li> <li>Fixed Mantle endpoint authentication failing with missing <code>x-api-key</code> header</li>…
<h2>What's changed</h2> <ul> <li>Added <code>--plugin-url <url></code> flag to fetch a plugin <code>.zip</code> archive from a URL for the current session</li> <li>Added <code>CLAUDE_CODE_FORCE_SYNC_OUTPUT=1</code> env var to force-enable synchronized output on terminals th…
<h2>What's changed</h2> <ul> <li>Bare <code>/color</code> (no args) now picks a random session color</li> <li><code>/mcp</code> now shows the tool count for connected servers and flags servers that connected with 0 tools</li> <li><code>--plugin-dir</code> now accepts <code>.zip</…
<h2>What's changed</h2> <ul> <li>The <code>/model</code> picker now lists models from your gateway's <code>/v1/models</code> endpoint when <code>ANTHROPIC_BASE_URL</code> points at an Anthropic-compatible gateway</li> <li> <ul> <li>Added <code>claude project purge [path]</code> t…
<h2>What's changed</h2> <ul> <li>Fixed OAuth authentication failing with a 401 retry loop when <code>CLAUDE_CODE_DISABLE_EXPERIMENTAL_BETAS=1</code> is set</li> </ul>
<h2>What's changed</h2> <ul> <li>Added <code>ANTHROPIC_BEDROCK_SERVICE_TIER</code> environment variable to select a Bedrock service tier (<code>default</code>, <code>flex</code>, or <code>priority</code>), sent as the <code>X-Amzn-Bedrock-Service-Tier</code> header</li> <li>Pasti…
<h2>What's changed</h2> <ul> <li>Added <code>alwaysLoad</code> option to MCP server config — when <code>true</code>, all tools from that server skip tool-search deferral and are always available</li> <li>Added <code>claude plugin prune</code> to remove orphaned auto-installed plu…
<h2>What's changed</h2> <ul> <li><code>/config</code> settings (theme, editor mode, verbose, etc.) now persist to <code>~/.claude/settings.json</code> and participate in project/local/policy override precedence</li> <li>Added <code>prUrlTemplate</code> setting to point the footer…
<h2>What's changed</h2> <ul> <li>Added vim visual mode (<code>v</code>) and visual-line mode (<code>V</code>) with selection, operators, and visual feedback</li> <li>Merged <code>/cost</code> and <code>/stats</code> into <code>/usage</code> — both remain as typing shortcuts that …
<h2>What's changed</h2> <ul> <li>Forked subagents can now be enabled on external builds by setting <code>CLAUDE_CODE_FORK_SUBAGENT=1</code></li> <li>Agent frontmatter <code>mcpServers</code> are now loaded for main-thread agent sessions via <code>--agent</code></li> <li>Improved …
<h2>What's changed</h2> <ul> <li><code>/resume</code> on large sessions is significantly faster (up to 67% on 40MB+ sessions) and handles sessions with many dead-fork entries more efficiently</li> <li>Faster MCP startup when multiple stdio servers are configured; <code>resources/…
<h2>0.95.0 (2026-04-14)</h2> <p>Full Changelog: <a href="https://github.com/anthropics/anthropic-sdk-python/compare/v0.94.1...v0.95.0">v0.94.1...v0.95.0</a></p> <h3>Features</h3> <ul> <li><strong>api:</strong> mark Sonnet and Opus 4 as deprecated (<a href="https://github.com/anth…
<h3>Internal/Other Changes</h3> <ul> <li>Switched test suite from <code>pytest-asyncio</code> to anyio's pytest plugin, running every async test under both asyncio and trio backends to catch backend-specific regressions in CI (<a class="issue-link js-issue-link" href="https://git…
Claude Agent SDK (Python) — Releases
TIER_1(SK)·github-actions[bot]·
<h3>Internal/Other Changes</h3> <ul> <li>Updated bundled Claude CLI to version 2.1.150</li> <li>Switched CI workflows from static API key to Workload Identity Federation for Claude authentication, using short-lived tokens instead of long-lived secrets (<a class="issue-link js-iss…
Claude Agent SDK (Python) — Releases
TIER_1(SK)·github-actions[bot]·
<h3>Breaking</h3> <ul> <li><strong>Breaking:</strong> MCP servers now connect in the background by default; sessions start immediately and slow servers report <code>status: "pending"</code> in <code>init</code> until ready. Set <code>MCP_CONNECTION_NONBLOCKING=0</code> to restore…
Claude Agent SDK (Python) — Releases
TIER_1(SK)·github-actions[bot]·
<h3>Bug Fixes</h3> <ul> <li><strong>Actionable error messages after error results</strong>: Replaced the generic <code>Command failed with exit code 1</code> exception raised after an error result with one carrying the result's actual error text (e.g. "Reached maximum number of t…
Claude Agent SDK (Python) — Releases
TIER_1(SK)·github-actions[bot]·
<h3>New Features</h3> <ul> <li><strong>API error status on result messages</strong>: Added <code>api_error_status: int | None</code> to <code>ResultMessage</code>, surfacing the HTTP status code (e.g. 429, 500, 529) from failing API calls. This provides a safe-to-log field for cl…
Claude Agent SDK (Python) — Releases
TIER_1(SK)·github-actions[bot]·
<h3>New Features</h3> <ul> <li><strong>Hook event streaming</strong>: Added <code>include_hook_events</code> option to <code>ClaudeAgentOptions</code>. When set, hook events (PreToolUse, PostToolUse, Stop, etc.) are emitted by the CLI and yielded from the message stream as <code>…
Claude Agent SDK (Python) — Releases
TIER_1(SK)·github-actions[bot]·
<h3>New Features</h3> <ul> <li><strong>Eager session store flushing</strong>: Added <code>session_store_flush</code> option to <code>ClaudeAgentOptions</code> (<code>"batched"</code> or <code>"eager"</code>). When set to <code>"eager"</code>, the transcript mirror delivers frames…
Claude Agent SDK (Python) — Releases
TIER_1(SK)·github-actions[bot]·
<h3>Documentation</h3> <ul> <li>Added docstrings to <code>ClaudeAgentOptions</code> fields for improved IDE autocompletion and inline documentation (<a class="issue-link js-issue-link" href="https://github.com/anthropics/claude-agent-sdk-python/pull/873">#873</a>)</li> </ul> <h3>…
Claude Agent SDK (Python) — Releases
TIER_1(SK)·github-actions[bot]·
<h3>Bug Fixes</h3> <ul> <li><strong>Trio compatibility restored</strong>: Fixed <code>RuntimeError: no running event loop</code> when using <code>ClaudeSDKClient</code> or <code>query()</code> under trio, a regression introduced in v0.1.51. Uses sniffio-based dispatch to select t…
Claude Agent SDK (Python) — Releases
TIER_1(SK)·github-actions[bot]·
<h3>New Features</h3> <ul> <li><strong>SessionStore adapter</strong>: Full SessionStore support at parity with the TypeScript SDK. Includes a <code>SessionStore</code> protocol with 5 methods (<code>append</code>, <code>load</code>, <code>list_sessions</code>, <code>delete</code>…
<h2>What's Changed</h2> <ul> <li>Add test that each model has unique identifiers (CORE-134) by <a class="user-mention notranslate" href="https://github.com/Talmaj">@Talmaj</a> in <a class="issue-link js-issue-link" href="https://github.com/Comfy-Org/ComfyUI/pull/13654">#13654</a>…
<h2>What's Changed</h2> <ul> <li>[Partner Nodes] new Flux2ImageNode and GrokImageEditNodeV2 nodes by <a class="user-mention notranslate" href="https://github.com/bigcat88">@bigcat88</a> in <a class="issue-link js-issue-link" href="https://github.com/Comfy-Org/ComfyUI/pull/13814">…
<h2>What's Changed</h2> <ul> <li>Read audio and video at the same time in video loader node. by <a class="user-mention notranslate" href="https://github.com/comfyanonymous">@comfyanonymous</a> in <a class="issue-link js-issue-link" href="https://github.com/Comfy-Org/ComfyUI/pull/…
<h1>Highlights</h1> <ul> <li> <p><strong>CUDA 13 + Torch 2.11</strong>: Default CUDA version moves to 13.0 across SGLang, sgl-kernel, and Docker images, and PyTorch is upgraded from 2.9 to 2.11 — modernizing the build matrix and unlocking newer kernels: <a class="issue-link js-is…
<h1>vLLM v0.20.1</h1> <p>This is a patch release on top of <code>v0.20.0</code> primarily focused on <strong>DeepSeek V4 stabilization and performance improvements</strong>, along with several important bug fixes.</p> <h3>DeepSeek V4</h3> <ul> <li>Base model support (<a class="is…
<h2>What's Changed</h2> <ul> <li>feat: SUPIR model support (CORE-17) by <a class="user-mention notranslate" href="https://github.com/kijai">@kijai</a> in <a class="issue-link js-issue-link" href="https://github.com/Comfy-Org/ComfyUI/pull/13250">#13250</a></li> <li>Some optimizati…
<p>This is a patch release on top of <code>v0.19.0</code> with Transformers v5.5.3 upgrade and bug fixes for Gemma4:</p> <ul> <li>Update to transformers v5 (<a class="issue-link js-issue-link" href="https://github.com/vllm-project/vllm/pull/30566">#30566</a>)</li> <li>[Bugfix] Fi…
<p><strong>Full Changelog</strong>: <a class="commit-link" href="https://github.com/sgl-project/sglang/compare/v0.5.10...v0.5.10.post1"><tt>v0.5.10...v0.5.10.post1</tt></a></p> <p>Bumps flashinfer from v0.6.7.post2 to v0.6.7.post3 to resolve an issue in its jit cubin downloader.<…
<h1>Highlights</h1> <ul> <li> <p><strong>Piecewise CUDA Graph Enabled by Default</strong>: Piecewise CUDA graph capture is now the default execution mode, reducing memory overhead and improving throughput for models with complex control flow patterns: <a class="issue-link js-issu…
<h1>vLLM v0.19.0</h1> <h2>Highlights</h2> <p>This release features 448 commits from 197 contributors (54 new)!</p> <ul> <li><strong>Gemma 4 support</strong>: Full Google Gemma 4 architecture support including MoE, multimodal, reasoning, and tool-use capabilities (<a class="issue-…
<h1>Highlights</h1> <ul> <li> <p><strong>Piecewise CUDA Graph Enabled by Default</strong>: Piecewise CUDA graph capture is now the default execution mode, reducing memory overhead and improving throughput for models with complex control flow patterns: <a class="issue-link js-issu…
<h2>2.26.0 (2026-03-05)</h2> <p>Full Changelog: <a href="https://github.com/openai/openai-python/compare/v2.25.0...v2.26.0">v2.25.0...v2.26.0</a></p> <h3>Features</h3> <ul> <li><strong>api:</strong> The GA ComputerTool now uses the CompuerTool class. The 'computer_use_preview' to…
<h1>Highlights</h1> <ul> <li> <p><strong>LoRA Weight Loading Overlap with Computation</strong>: Overlap LoRA weight loading with computation during inference, reducing TTFT by ~78% and TPOT by ~34.88% on large adaptors: <a class="issue-link js-issue-link" href="https://github.com…
<h1>Highlights</h1> <ul> <li>Up to 1.5x faster across the board for all major diffusion models <a href="https://lmsys.org/blog/2026-01-16-sglang-diffusion/" rel="nofollow">https://lmsys.org/blog/2026-01-16-sglang-diffusion/</a></li> <li>Close to linear scaling with chunked pipeli…
dev.to — Claude Code tag
TIER_1Deutsch(DE)·JessYT·
<h1> Claude Code unlocked background again — Fast mode is now 4.7 </h1> <p>Bottom line first. <strong>Claude Code's Fast mode default auto-switched to Opus 4.7.</strong> The same changelog batch also shipped 8 new <code>claude agents</code> CLI flags, an option to turn off worktr…
<h2> Introduction </h2> <p>Everyone knows that the MAX plan lets you "use more."</p> <p>That's not what I want to know. <strong>What actually changes with MAX?</strong> When I looked into it, I found that some features often assumed to be "MAX-exclusive" were not, while other dif…
dev.to — Claude Code tag
TIER_1English(EN)·Jula Markova·
<p>I spent a session with Claude Code (Opus 4.7) doing something odd. Instead of giving it tasks, I asked it to reflect on its own thinking. Not what it knows. How it <em>operates</em>.</p> <p>What came back was specific enough to be useful. </p> <p>One conversation = One experim…
dev.to — Claude Code tag
TIER_1English(EN)·Alaric·
<p>I've been using Claude Code daily for the past few months. The output quality is great, but the API bill at the end of the month was painful. After some experimentation, I found a few habits that consistently cut my token consumption by 25–35% without sacrificing code quality.…
dev.to — Claude Code tag
TIER_1Nederlands(NL)·speedy_devv·
<p>Your Claude Code bill went up twice in 30 days. You felt it on the invoice before you understood why.</p> <p>On June 15, 2026, Anthropic moves Agent SDK, <code>claude -p</code>, and Claude Code GitHub Actions onto a separate metered credit pool that does not roll over. Once th…
Medium — Claude tag
TIER_1English(EN)·Bhavin Mecwan·
<h1> Claude's "Go to Sleep" Bug Explained: Why It Happens and How to Keep Coding </h1> <p><strong>TL;DR:</strong> Claude has been telling developers to "get some rest" mid-coding session, even at 9 a.m., and the behaviour has been growing for months. Anthropic staffer Sam McAllis…
<p>I tested cheap vs expensive LLMs across 3 real agent tasks. The cheap model won every time.<br /> Everyone tells you to use the best model you can afford. I built a toolkit to test that assumption with data, and the results surprised me.<br /> Over the past few weeks, I built …
<!-- SC_OFF --><div class="md"><p>Backstory: when the "Build with Opus 4.7" hackathon was announced, I had the thesis but not the project. I was obsessing over the tokenomics of agents; how to make your tokens go further, and shared that angle in the application even wi…
