The AI Toolchain — issue -018, August 1, 2026

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

Cline SDK v0.0.68 adds getShellInvocation() for UTF-8 PowerShell stdin and improves model catalog display names

$ git clone --branch sdk/sdk/v0.0.68 https://github.com/cline/cline.git
# already have the repo? check out this version:
$ git checkout sdk/sdk/v0.0.68

• ›Adds getShellInvocation() to replace the now-deprecated getShellArgs(), routing PowerShell commands over UTF-8 stdin so non-ASCII characters survive the active code page and long commands are no longer capped by the Windows command-line length limit; stdin write failures surface as a command error instead of hanging.
• ›Exports finish-reason and auth-error helpers for describing agent errors.
• ›Provider errors forwarded through the Vercel AI Gateway now surface the real upstream message instead of a raw Zod issue dump or [object Object].

Qwen Code v0.21.3 adds fork profiles, display_image tool, /review test-plan validation, and background workflow runs.

$ git clone --branch v0.21.3 https://github.com/QwenLM/qwen-code.git
# already have the repo? check out this version:
$ git checkout v0.21.3

$ /model --compaction qwen2.5-72b-instruct

{
  "generationConfig": {
    "cacheRetention": "1h"
  }
}

skills:
  disabledLevels:
    - bundled
    - project

• ›Adds generationConfig.cacheRetention settings to support Anthropic's extended 1-hour prompt-cache tier alongside the default 5-minute retention.
• ›Adds skills.disabledLevels setting to hide bundled, project, user, or extension skills without affecting host-provided skills.
• ›Adds --compaction flag to the /model command to configure a dedicated model for chat compression.
• ›Enables POST /session to accept and validate a caller-supplied sessionId instead of silently ignoring it.
• ›Adds a SessionDelete hook event triggered after successfully deleting a historical session via /delete or ACP.

KoboldCpp v1.118.1 adds image-gen abort support, multi-language Qwen3TTS, --jinjathink flag, and int8 convrot image model loading.

$ git clone --branch v1.118.1 https://github.com/LostRuins/koboldcpp.git
# already have the repo? check out this version:
$ git checkout v1.118.1

$ koboldcpp --model mymodel.gguf --jinjathink

• ›New --jinjathink flag automatically enables Jinja templating when specified.
• ›Image generation can now be aborted mid-generation by closing the connection; disable this behavior per-request with the keep_image_gen_on_disconnect request field.
• ›Adds support for loading int8 convrot image models (z-image models are excluded from image reference support).
• ›Adds multiple language support for Qwen3TTS voices, available via the XTTS endpoint.
• ›Adds support for Qwen3TTS aliases for all voices.

llama-cli now persists reasoning_content in chat history so --reasoning-preserve can re-inject prior thoughts on later turns.

$ git clone --branch b10219 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b10219

• ›Stores reasoning_content in the chat message history within llama-cli, enabling --reasoning-preserve to re-inject prior model thoughts on subsequent conversation turns.

$ git clone --branch b10218 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b10218

$ git clone --branch b10217 https://github.com/ggml-org/llama.cpp.git
# already have the repo? check out this version:
$ git checkout b10217

OpenAI Agents SDK v0.19.2 exposes original callables through wrapped functions for easier introspection.

$ git clone --branch v0.19.2 https://github.com/openai/openai-agents-python.git
# already have the repo? check out this version:
$ git checkout v0.19.2

• ›Exposes the original callable via a wrapped attribute on wrapped functions, enabling runtime introspection of the underlying function.

PydanticAI v2.22.0 adds RunContext.is_tool_available, MCP task-skipping via prefer_tasks, and Gemini VALIDATED tool mode by default.

$ git clone --branch v2.22.0 https://github.com/pydantic/pydantic-ai.git
# already have the repo? check out this version:
$ git checkout v2.22.0

from pydantic_ai.mcp import MCPToolset

toolset = MCPToolset(
    server_url='http://localhost:3000',
    prefer_tasks=False  # skip optional MCP tasks rather than blocking
)

• ›Adds RunContext.is_tool_available method, letting tool code check at runtime whether another named tool is accessible in the current agent context.
• ›Adds prefer_tasks parameter to MCPToolset clients, allowing optional MCP tasks to be skipped when the runtime does not support them.
• ›Adds configurable max_retries to ToolSearchToolset, giving control over how many times a tool-search lookup is retried on failure.
• ›Enables Gemini VALIDATED tool mode by default on supported models, improving structured tool-call reliability without manual configuration.
• ›Sends mid-conversation system prompts as native system messages on Anthropic, aligning prompt delivery with Anthropic's native message format.

Phoenix v19.13.0 adds annotations to span exports and lifts the experiment metric chart selection limit.

$ git clone --branch arize-phoenix-v19.13.0 https://github.com/Arize-ai/phoenix.git
# already have the repo? check out this version:
$ git checkout arize-phoenix-v19.13.0

• ›Span and trace annotations are now included when exporting spans, enriching exported data with human and automated feedback signals.
• ›Removes the metric chart selection limit in the experiments view, allowing unlimited metrics to be charted simultaneously.