The daily firehose — everything the toolchain shipped today, already filtered.
// HOW THIS ISSUE IS MADE
We read every release from the 174 tools on our watchlist at the source — GitHub and GitLab release notes, vendor release pages and changelogs, project blogs and feeds, vendor press releases, and the source code behind the tag. Bug-fix-only releases and non-product newsroom noise are dropped; what's left is summarized down to the new capability, how to try it, and any screenshots or videos the release itself published. Every entry links to the sources it was built from.
Claude Code is an agentic coding tool that lives in your terminal, understands your codebase, and helps you code faster by executing routine tasks, explaining complex code, and handling git workflows - all through natural language commands.
Claude Code v2.1.198 ships background subagents by default, auto-PR on worktree completion, /dataviz skill, and AWS upstream gateway support.
└──▷ GET THIS VERSION
$ git clone --branch v2.1.198 https://github.com/anthropics/claude-code.git
# already have the repo? check out this version:$ git checkout v2.1.198
└──▷ TRY IT
Receive a desktop notification when a background agent finishes or needs input — useful for long-running tasks you want to monitor hands-off.
$ claude agents
›Subagents now run in the background by default, letting Claude continue working and receive notifications when they finish.
›Claude in Chrome is now generally available.
›Adds agent_needs_input and agent_completed notification hooks to claude agents via the Notification hook.
›New /dataviz skill for chart and dashboard design guidance with a runnable color-palette validator.
›Gateway adds Claude Platform on AWS (anthropicAws) as an upstream provider; model-not-found responses now advance the failover chain.
+10 moreshow less
›Background agents launched from claude agents now automatically commit, push, and open a draft PR when they finish code work in a worktree.
›The built-in Explore agent now inherits the main session's model (capped at opus) instead of running on haiku.
›Subagents and context compaction now inherit the session's extended thinking configuration, improving output quality on delegated tasks.
›Improved focus mode: subagents launched in a turn appear in its activity summary, and completed background notifications fold into a single count.
›Syntax highlighting accuracy improved in code blocks, diffs, and file previews by upgrading to highlight.js 11.
›Keyboard shortcut hints now show opt/cmd instead of alt/super when connected from a Mac over SSH.
›API retry UX now shows the error reason after the second attempt, and displays a status page link when the API is overloaded.
›/login now opens the sign-in dialog from the claude agents view.
›Subagents now treat messages from the agent that launched them as normal task direction.
›Removed the /agents wizard; subagents are now created by asking Claude or editing .claude/agents/ directly.
└──▷ BREAKING ON UPGRADE
!The /agents wizard has been removed; subagent creation must now be done by asking Claude directly or editing .claude/agents/ manually.
!claude --bg combined with --print/-p is now rejected up front instead of silently creating an unattachable session.
GitHub Copilot CLI v1.0.68 adds kimi-k2.7-code model support, inline tab-completion aliases, plan budget visibility, and session management from the agents screen.
└──▷ GET THIS VERSION
$ git clone --branch v1.0.68 https://github.com/github/copilot-cli.git
# already have the repo? check out this version:$ git checkout v1.0.68
└──▷ TRY IT
Check plan budget consumption mid-session without leaving the CLI.
›Shows plan budget details in the statusline and /usage for supported plans.
›Browse, resume, and switch between sessions from the agents screen.
›Focused field in the /mcp config form is marked with a '❯ ' chevron, not color alone.
+8 moreshow less
›Keeps IDE tools available during transient IDE disconnects, returning a clear error while disconnected and recovering automatically when the IDE reconnects.
›Tells the agent when the working directory changes between turns so it uses the new directory for commands and relative paths.
›Shows descriptions for slash-command input choices.
›Custom agents keep their tool filters in nested subagents.
›Adds git_panel.entry_primary_click_action setting to control default click behavior for Git Panel files.
›Adds markdown_preview.limit_content_width and markdown_preview.max_width settings for constraining and centering Markdown preview content.
›Adds shell completion generation for the zed CLI via zed --completions <SHELL>.
›Adds git_panel.sort_by (values: 'path' or 'name') and git_panel.group_by (values: 'none' or 'status') settings to the Git Panel View Options menu.
›Adds a text finder picker with live preview, opened via the text finder: toggle action in the command palette, as an alternative project search UI with shared searches.
+13 moreshow less
›Adds previews to the file finder, displayed to the right or below results depending on available space.
›Adds draggable resizing to picker modals.
›Adds in-thread search to the Agent Panel via Ctrl/Cmd+F.
›Adds an entry to the Agent Panel options menu to quickly add a remote MCP server.
›Adds ACP support for embedded resources in tool calls.
›Adds a Git Panel View Options menu for switching between list/tree view, sorting by path or name, and grouping by status; Project Diff now follows the same file ordering as the Git Panel.
›Adds a Git blame toggle to the gutter's context menu.
›Improves the Git Panel changes header with a split button keeping Stage All / Unstage All as the primary action and grouping related change actions in a dropdown.
›Adds Helix-mode debugger keybindings under space shift-g for starting, restarting, continuing, pausing, stepping, stopping, toggling breakpoints, and editing log breakpoints.
›Adds paragraph navigation in terminal Vim mode with shift-{ and shift-}.
›Adds GLM 5.2 and Kimi K2.7 Code to OpenCode Go, and makes DeepSeek V4 Pro available for OpenCode Zen.
›Improves Git Graph search to find commits by abbreviated or full hash.
›Includes the thread title in Agent Panel notifications when Zed is not focused.
└──▷ BREAKING ON UPGRADE
!The git_panel.sort_by_path setting is replaced by git_panel.sort_by ('path' or 'name') and git_panel.group_by ('none' or 'status'); existing configs using git_panel.sort_by_path will break.
!MiniMax M3 Free is removed and deprecated OpenCode models GLM 5, Kimi K2.5, and MiniMax M2.5 are limited to Zen only.
Hermes v2026.7.1 adds MoA as a selectable model, /learn, /journey, /prompt, Vertex AI OAuth2, background subagent fan-out, and completion contracts for /goal.
└──▷ GET THIS VERSION
$ git clone --branch v2026.7.1 https://github.com/NousResearch/hermes-agent.git
# already have the repo? check out this version:$ git checkout v2026.7.1
└──▷ TRY IT
Teach Hermes a custom triage workflow from your runbook URL so it's available as a reusable skill in future sessions.
Compose a detailed, multi-line threat-model prompt in your editor instead of fighting a single-line input.
$ /prompt
›Adds Mixture-of-Agents (MoA) presets as selectable virtual models under a moa provider, available in CLI, TUI, desktop, and gateway model pickers.
›Adds /moa as a one-shot command that runs a single prompt through the default MoA preset, then restores the previous model.
›Streams each reference model's reasoning as labelled blocks before the aggregator's live-streamed final answer.
›Adds /learn <anything> to distill a reusable skill from a directory, URL, or recent workflow and write it automatically per CONTRIBUTING.md standards.
›Adds /journey to the CLI and TUI: a browsable, editable timeline of all memories and skills the agent has accumulated.
+10 moreshow less
›Adds a desktop memory graph — a playable radial timeline of memories and skills that can be pruned from the UI.
›Adds completion contracts to /goal: define what 'done' looks like and the standing-goal loop judges completion against evidence, not model assertion.
›Adds a pre_verify hook for wiring custom checks into the goal-completion verification loop.
›Enables delegate_task to fan out multiple subagents in the background, returning a single consolidated result when all finish.
›Adds first-class coding Projects to the desktop app, including a sidebar, coding rail, review pane, git worktree management, and a project → repo → lane model.
›Adds scale-to-zero support and drain coordination to the gateway, allowing idle instances to go dormant and resume without dropping in-flight conversations.
›Adds Google Vertex AI as a first-class provider, with automatic OAuth2 token minting and refresh from a service-account JSON or Application Default Credentials.
›Adds /prompt command that opens $EDITOR for composing multi-line prompts in markdown before queuing them as the next message.
›Routes the post-turn self-improvement fork to an auxiliary model with digested context, reducing cost of background memory/skill saving.
›Hardens MCP-config persistence attack surface, blocks cron base_url credential-exfiltration overrides, adds non-reusable sentinel for prefix secrets in file reads, redacts Slack xapp- tokens, enforces a browser cloud-metadata floor on every backend, and applies an aiohttp CVE floor across lazy messaging paths.
OpenCL backend gains support for loading precompiled binary kernels from an external library, including q8_0, q4_0, q4_1, q4_k, and MoE MX-FP4 GEMM kernels.
└──▷ GET THIS VERSION
$ git clone --branch b9859 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b9859
›Adds OpenCL support for loading precompiled binary kernels (including gemm_moe_mxfp4_f32_ns, q8_0 GEMM, q4_0/q4_1/q4_k MoE GEMM) from an external kernel library at runtime, enabling vendor-optimized kernel delivery without recompilation.
Phoenix v17.15.0 adds Claude Sonnet 5 to the Anthropic and AWS Bedrock playground providers.
└──▷ GET THIS VERSION
$ git clone --branch arize-phoenix-v17.15.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:$ git checkout arize-phoenix-v17.15.0
›Adds Claude Sonnet 5 as a selectable model in the Playground for both the Anthropic and AWS (Bedrock) providers.
THE AI TOOLCHAIN