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.
›Adds Discord voice-follow: voice sessions now track configured users into channels with allowed-channel checks, multi-user handoff, bounded reconciliation, and DAVE recovery preservation.
›Adds automatic injection of bounded IDENTITY.md, USER.md, and SOUL.md profile context into realtime voice session instructions by default, configurable via voice.realtime.bootstrapContextFiles.
›Adds bundled Policy plugin for policy-backed channel conformance checks, doctor lint findings, and opt-in workspace repair.
›Adds agents.list[].experimental.localModelLean config option to enable lean local-model mode per agent instead of globally.
›Adds device-code OAuth login for xAI provider, enabling authorization from remote and headless environments without a localhost browser callback.
+1 moreshow less
›Adds provider-level params.provider routing policy support for OpenRouter requests, with model and agent params able to override defaults.
└──▷ BREAKING ON UPGRADE
!The old cat SKILL.md && printf ... && <skill-wrapper> allowlist compatibility path for exec approvals is removed; skill files must now be loaded with the read tool and only the real skill executable is auto-allowed.
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 v2.1.147 adds a new /code-review command with inline GitHub PR comments and pinned background session persistence.
└──▷ GET THIS VERSION
$ git clone --branch v2.1.147 https://github.com/anthropics/claude-code.git
# already have the repo? check out this version:$ git checkout v2.1.147
└──▷ TRY IT
Run a high-effort correctness review on your current code and post findings as inline comments on an open GitHub PR.
$ /code-review high --comment
›New /code-review command (replaces /simplify) reports correctness bugs at a chosen effort level and can post findings as inline GitHub PR comments via --comment.
›Pinned background sessions (Ctrl+T in claude agents) now stay alive when idle, restart in place to apply updates, and are only shed under memory pressure after non-pinned sessions.
└──▷ BREAKING ON UPGRADE
!The /simplify command is renamed to /code-review and its cleanup-and-fix behavior has been removed; scripts or workflows invoking /simplify will no longer work.
$ git clone --branch rust-v0.133.0 https://github.com/openai/codex.git
# already have the repo? check out this version:$ git checkout rust-v0.133.0
›Enables Goals by default with dedicated storage and cross-turn progress tracking.
›Permission profiles gain a list API, inheritance resolution, managed requirements.toml support, runtime refresh, and deeper Windows sandbox integration.
›Plugin list output is now tabulated and marketplace-aware, showing installed versions, visible marketplace roots, and remote collection support.
›Extensions can now observe SubagentStart/SubagentStop lifecycle hooks, tool execution events, turn-start metadata, and async approval/turn processing.
›Adds a /btw alias for the side slash command.
+5 moreshow less
›Adds turn_id and truncation_policy fields to extension tool calls.
›Adds a standalone web-search API client.
›Adds MITM hook config model and wires MITM hooks into runtime enforcement with named permissions config.
›Adds thread/settings/update app-server API for syncing TUI thread settings.
Qwen Code v0.16.0 adds Auto approval mode, /goal, /stuck, /diff, worktree isolation, ModelScope provider, and a new prompt hook type.
└──▷ GET THIS VERSION
$ git clone --branch v0.16.0 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:$ git checkout v0.16.0
└──▷ TRY IT
Set a session goal so Qwen Code judges each turn against it and continues autonomously until done — or fails fast if it detects the goal is impossible.
$ /goal Refactor all usages of deprecated `fs.exists` to `fs.stat` across the repo
Configure a custom directory for Plan Mode plans instead of the default location.
json
{
"plansDirectory": "./docs/plans"
}
Run memory diagnostics to spot heap pressure or context bloat before a long agentic session.
$ /doctor
›Adds --auto completion flag to /rename with argument hint support.
›Adds plansDirectory config key to make the directory used by Plan Mode configurable.
›Adds TodoCreated and TodoCompleted hooks for todo lifecycle events.
›Adds a new prompt hook type with LLM evaluation support.
›Adds stop hook blocking capability.
+26 moreshow less
›Adds EnterWorktree and ExitWorktree tools for generic worktree support with agent isolation, plus Phase C session persistence, hooksPath, Footer, WorktreeExitDialog, and three-mode --resume restore.
›Adds /goal command — session-scoped goal setting with judge-driven turn continuation and active goal stream events, including non-interactive goal support; impossible goals are now detected and failed automatically.
›Adds /stuck diagnostic skill for frozen or stalled sessions.
›Adds /doctor command with baseline memory diagnostics and structured memory diagnostics JSON output.
›Adds /diff command with per-turn interactive dialog.
›Adds file restoration support to the /rewind command.
›Adds custom output directory support for /export.
›Adds fork-session --resume flag.
›Adds ModelScope as a built-in third-party API provider.
›Adds Ctrl+P / Ctrl+N readline keybindings for history and selection navigation.
›Adds built-in status line presets with an interactive selection dialog.
›Adds session path status command.
›Adds atomicWriteFile generic write primitive wired into the Write and Edit tools.
›Adds a /demo debug page for the qwen serve daemon.
›Adds Auto approval mode with an LLM classifier.
›Adds image and video support for Qwen3.6-35B-A3B quantized variants.
›Enables progressive MCP availability — MCP no longer blocks the first user input.
›Injects git status into the system prompt and refines Explore/git-log guidance.
›Uses fastModel for the Explore subagent.
›Extends cross-auth fast models to agents.
›Strips inline media before chat compaction summary to reduce token overhead.
›Applies /language output change to the running session without requiring a restart.
›Warns users when rewind is disabled in IDE mode.
›Supports batch deletion of sessions via /delete.
›Unified span creation paths for hierarchical trace trees with interaction span and detailed sensitive attributes.
›Code-splits lowlight syntax highlighting to cut startup V8 parse cost.
llama.cpp b9270 adds native HybridDNATokenizer support for HuggingFaceBio Carbon models via a new LLAMA_VOCAB_TYPE_HYBRIDDNA vocab type.
└──▷ GET THIS VERSION
$ git clone --branch b9270 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b9270
›Adds LLAMA_VOCAB_TYPE_HYBRIDDNA = 7 in include/llama.h as a first-class vocab type (peer to SPM/WPM/UGM/RWKV/PLAMO2) for the HybridDNATokenizer used by HuggingFaceBio/Carbon-{500M,3B,8B}.
›Adds llm_tokenizer_hybriddna and its session in src/llama-vocab.cpp, routing text through a DNA-aware splitter that chunks sequences inside <dna>...</dna> regions into fixed 6-mers (right-padded with A), maps non-ACGT bases to <oov>, and delegates non-DNA text to an embedded BPE tokenizer.
›Adds _set_vocab_hybriddna in conversion/base.py that writes tokenizer.ggml.model = 'hybriddna' during GGUF conversion, dispatched from conversion/llama.py when tokenizer_config.json["tokenizer_class"] == "HybridDNATokenizer".
└──▷ BREAKING ON UPGRADE
!The short-lived LLAMA_VOCAB_PRE_TYPE_CARBON (introduced during this same development branch) is removed from src/llama-vocab.h in favour of LLAMA_VOCAB_TYPE_HYBRIDDNA; any code referencing LLAMA_VOCAB_PRE_TYPE_CARBON will fail to compile.
llama.cpp b9263 merges HunyuanOCR into the HunyuanVL projector path, restoring correct vision precision.
└──▷ GET THIS VERSION
$ git clone --branch b9263 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:$ git checkout b9263
›Merges HunyuanOCR into the HUNYUANVL projector and HUNYUAN_VL text architecture, consolidating OCR and vision-language into a single code path with the +0.1 bilinear sampler applied correctly.
Phoenix v16 adds sandboxed Code Evaluators and provider-native web search/fetch for agents.
└──▷ GET THIS VERSION
$ git clone --branch arize-phoenix-v16.0.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:$ git checkout arize-phoenix-v16.0.0
›Adds sandboxed Code Evaluators: write a Python or TypeScript evaluate() function directly in the Phoenix UI and Phoenix runs it server-side, recording labels and scores as annotations on every experiment run — no SDK, no local runtime, no deploy step required.
›Code Evaluators support composite scoring (blending LLM judgment with deterministic rules into a weighted metric), embedding-based evaluation (cosine similarity over embeddings), and LLM juries (polling multiple models into a weighted consensus verdict).
›Enables provider-native web search and fetch tools for agents when the underlying provider supports them.
└──▷ BREAKING ON UPGRADE
!Sandboxing and Code Evaluators introduce a breaking change — see MIGRATION.md at https://github.com/Arize-ai/phoenix/blob/main/MIGRATION.md for upgrade steps.
Phoenix v15.12.0 stamps tool execution environment on tool-call provider metadata and adds Enter-key submission for elicitation responses.
└──▷ GET THIS VERSION
$ git clone --branch arize-phoenix-v15.12.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:$ git checkout arize-phoenix-v15.12.0
›Stamps tool execution environment on tool-call provider metadata for agent spans, making runtime context visible in traces.
›Enables submitting elicitation responses with the Enter key in the UI, reducing friction during annotation workflows.
LanceDB python-v0.33.0-beta.0 aligns to_pandas kwargs and adds Node.js table rename and add-progress features.
└──▷ GET THIS VERSION
$ git clone --branch python-v0.33.0-beta.0 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:$ git checkout python-v0.33.0-beta.0
›Aligns to_pandas to accept pandas keyword arguments directly in the Python client.
›Adds renameTable method on Connection in the Node.js client to rename tables.
›Adds a progress callback to Table.add in the Node.js client for tracking ingestion progress.
└──▷ BREAKING ON UPGRADE
!Nested field paths in native index creation are now handled differently; existing setups relying on the prior path format for nested fields may break on upgrade.
LanceDB v0.30.0-beta.0 adds progress reporting to Table.add and renameTable on Connection in Node.js, plus pandas kwarg alignment in Python.
└──▷ GET THIS VERSION
$ git clone --branch v0.30.0-beta.0 https://github.com/lancedb/lancedb.git
# already have the repo? check out this version:$ git checkout v0.30.0-beta.0
›Adds renameTable method on Connection in the Node.js SDK to rename tables in place.
›Adds progress callback support to Table.add in the Node.js SDK to track insertion progress.
›Aligns to_pandas pandas kwargs in the Python SDK so all upstream pandas keyword arguments are passed through.
└──▷ BREAKING ON UPGRADE
!Nested field paths in native index creation now use canonical nested index path format, which may change how existing nested-field indexes are addressed or referenced.
THE AI TOOLCHAIN