The AI Toolchain — issue -019, July 31, 2026

Autonomous coding agent as an SDK, IDE extension, or CLI assistant.

Cline v4.1.2 surfaces the active extension variant ('Legacy' or 'Next') in the settings About page.

$ git clone --branch v4.1.2 https://github.com/cline/cline.git
# already have the repo? check out this version:
$ git checkout v4.1.2

• ›Displays which extension variant is active — 'Legacy' or 'Next' — next to the version number in the settings About page for both bundles of the combined A/B rollout package.

$ git clone --branch cli-v3.0.48 https://github.com/cline/cline.git
# already have the repo? check out this version:
$ git checkout cli-v3.0.48

{
  "mcpServers": {
    "my-slow-server": {
      "command": "npx",
      "args": ["-y", "my-mcp-server"],
      "timeout": 30000
    }
  }
}

$ cline history

$ git clone --branch sdk/sdk/v0.0.67 https://github.com/cline/cline.git
# already have the repo? check out this version:
$ git checkout sdk/sdk/v0.0.67

{
  "mcpServers": {
    "my-server": {
      "command": "npx",
      "args": ["my-mcp-server"],
      "timeout": 300
    }
  }
}

const messages = await ClineCore.readLiveMessages(sessionId);

Glamourous agentic coding for all

Crush v0.88.0 introduces bash-based .crushrc config, permissions deny, option reset, and improved AWS Bedrock SSO re-auth.

$ git clone --branch v0.88.0 https://github.com/charmbracelet/crush.git
# already have the repo? check out this version:
$ git checkout v0.88.0

# ~/.crushrc
provider add ollama --type ollama --base-url "http://localhost:11434/v1"
model add ollama/llama3.3 --name "Llama 3.3" --context-window 128000
permissions allow view edit
mcp add github \
  --type http \
  --url "https://api.githubcopilot.com/mcp/" \
  --header Authorization "Bearer $GITHUB_TOKEN"
source "$XDG_CONFIG_HOME/local-overrides.sh"

# ~/.crushrc
option reset skill-path
if [[ $HOSTNAME == "workstation" ]]; then
    option skill-path "$HOME/work-skills"
fi

• ›Adds .crushrc bash-based configuration file, replacing and coexisting with the legacy JSON config, supporting provider add, model add, permissions allow, permissions deny, mcp add, option, source, and conditional shell logic for machine-specific overrides.
• ›Adds permissions deny command in .crushrc to explicitly hide specific tools from the agent.
• ›Adds option reset command in .crushrc to wipe shell config list options.
• ›Adds provider add and model add as verb-first shell config builtins for registering providers and models from .crushrc.
• ›Adds mcp add and mcp remove as verb-first shell config commands configurable from .crushrc, including --type, --url, and --header flags for HTTP MCP servers.

Qwen Code v0.21.2 adds Auto Recall, review run, fork_tools allowlist, GitLab polling, DingTalk cards, and a security.allowPrivateNetworkHooks setting.

$ git clone --branch v0.21.2 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:
$ git checkout v0.21.2

$ qwen review run

security:
  allowPrivateNetworkHooks: true

fork_tools:
  - read_file
  - grep
  - bash

• ›Adds security.allowPrivateNetworkHooks setting to allow HTTP hooks to reach private network addresses in trusted platform-managed environments.
• ›Adds reasonFilter setting for GitHub channels to skip unwanted notification reasons and prevent unintended agent actions.
• ›Introduces fork_tools allowlist on fork-type subagents to restrict execution to specific canonical tool names or MCP server patterns.
• ›Adds review run subcommand for headless, machine-readable review verdicts.
• ›Adds qwen review test-efficacy command that detects unprotected safety statements using statement-level mutation probes.

Framework for orchestrating role-playing, autonomous AI agents. By fostering collaborative intelligence, CrewAI empowers agents to work together seamlessly, tackling complex tasks.

CrewAI 1.15.10 adds skill usage event collection for agent observability.

$ git clone --branch 1.15.10 https://github.com/crewAIInc/crewAI.git
# already have the repo? check out this version:
$ git checkout 1.15.10

• ›Collects skill usage events, enabling observability into which agent skills are invoked during crew runs.

An MCP that lets AI tools securely connect to your infrastructure, write IaaS code, debug issues, and assist during incidents - without risking production stability. Built for security teams to approve and infrastructure teams to experience like magic.

emisar v0.36.0 adds staged runbooks with typed inputs and parallel execution, MCP 2026-07-28 routing support, and hardened OIDC/SCIM identity controls.

$ git clone --branch v0.36.0 https://github.com/AndrewDryga/emisar.git
# already have the repo? check out this version:
$ git checkout v0.36.0

• ›Adds staged runbooks where typed inputs bind once, stages run sequentially or in parallel against selected runner groups, and steps can extract named outputs, test success conditions, and wait within explicit bounds.
• ›Adds a single pre-execution approval that freezes the complete runbook execution plan before work begins; Canonical JSON uses the same definition shape across the console and MCP.
• ›Adopts the 2026-07-28 MCP routing headers and dual-era endpoint in the bridge and portal, including OAuth Client ID Metadata Documents.
• ›Expands the action catalog to 91 packs and 1,386 actions, adding bounded diagnostics for GCP, Pure FlashArray, Terraform, Nomad, OIDC/JWKS, nftables, TCP, and Docker Compose.
• ›Adds a unified bounded address policy covering OIDC discovery, JWKS refresh, and redirects that rechecks the response actually received; changing an issuer now clears credentials tied to the old one.

$ git clone --branch mcp-v0.5.0 https://github.com/AndrewDryga/emisar.git
# already have the repo? check out this version:
$ git checkout mcp-v0.5.0

llama.cpp b10216 adds full GGML_OP_POOL_1D support to the Vulkan backend via a new compute shader.

$ git clone --branch b10216 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b10216

• ›Adds GGML_OP_POOL_1D support to the Vulkan backend, including vk_op_pool1d_push_constants, pipeline_pool1d_f32, and the pool1d.comp compute shader, enabling 1D pooling operations on Vulkan-accelerated hardware.

$ git clone --branch b10208 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b10208

$ git clone --branch b10204 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b10204

$ git clone --branch b10203 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b10203

vMLX - JANGTQ Uber Compressed MLX Models - L2 Disk Cache (survives restart) + L1 Paged (super fast ttft) + Hybrid SSM Scheduler + Cont Batching + etc!

vMLX 1.6.19 adds native-MTP diagnostics, model-owned sampler hydration, and macOS Tahoe packaging.

$ git clone --branch v1.6.19 https://github.com/jjang-ai/vmlx.git
# already have the repo? check out this version:
$ git checkout v1.6.19

• ›Native-MTP health and profiling now expose bounded cache-lifecycle, acceptance, and phase-timing telemetry for Qwen MTP models.
• ›Ollama requests now preserve an explicitly supplied top_k, including 0, instead of silently dropping it during request translation.
• ›Automatic prompt limits are now derived from the configuration of the model that is actually loaded, bounded by the model's declared context ceiling and current memory budget.
• ›Separate Apple-silicon downloads are now provided for macOS Tahoe and for Sequoia-compatible systems.
• ›New and restored Electron chats now hydrate temperature, Top P, Top K, repetition penalty, and output limits from the selected model bundle.

Arize Phoenix v19.12.0 adds a Harbor proof-of-concept integration for evals.

$ git clone --branch arize-phoenix-v19.12.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:
$ git checkout arize-phoenix-v19.12.0

• ›Adds Harbor proof-of-concept integration to the evals module.

Langfuse v4.2.0 adds MCP telemetry, metrics discovery metadata, and automatic overflow of oversized trace fields to media storage.

$ git clone --branch v4.2.0 https://github.com/langfuse/langfuse.git
# already have the repo? check out this version:
$ git checkout v4.2.0

• ›Adds MCP tool outcome telemetry to Datadog for canonical observability of MCP tool calls.
• ›Exposes metrics query discovery metadata via MCP, enabling clients to enumerate available metrics queries.
• ›Exposes observation scope requirements via MCP so clients can understand required scoping for observations.
• ›Automatically moves oversized direct-event fields in trace observations to media storage rather than dropping or truncating them.
• ›Presents oversized fields as attachments in the UI instead of raw inline content.

Weave v0.53.4 adds custom runtime registration APIs, Azure workload identity support, and expanded Claude Agent SDK tracing including image prompts.

$ git clone --branch v0.53.4 https://github.com/wandb/weave.git
# already have the repo? check out this version:
$ git checkout v0.53.4

• ›Adds a custom runtime registration API (Python) and a custom runtime SDK wrapper — enabling programmatic registration of custom execution environments.
• ›Adds a custom runtime TypeScript SDK wrapper, bringing the same custom runtime capability to the TypeScript SDK.
• ›Supports Azure workload identity for authentication, removing the need for explicit credential secrets in Azure-hosted deployments.
• ›Traces Claude Agent SDK image prompts, extending multimodal visibility into Claude Agent workflows.
• ›Records turn output messages in agent traces, capturing full multi-turn conversation output for replay and debugging.

$ git clone --branch v0.53.3 https://github.com/wandb/weave.git
# already have the repo? check out this version:
$ git checkout v0.53.3