v1.5.9
中文
本版本重构本地 AI Agent CLI 的发现与选择机制,并加入 Codex GPT-5.6 模型支持。
- Codex、Claude Code、CodeWhale 统一使用跨平台 CLI 发现、能力探测和缓存机制,覆盖 Windows、macOS、Linux 与 WSL。
- 自动模式优先使用
PATH中第一个兼容版本,再检查原生安装、包管理器位置和受支持的旧路径;检测选择和手动路径模式严格锁定,不再静默回退。 - GUI、WebUI 和命令行新增 Agent CLI 候选列表、实际路径、版本、来源、兼容状态、测试与重新扫描入口。
- 新增
ngawolf agent-cli list|test|rescan管理命令,ngawolf check会显示实际解析的 Agent CLI。 - 兼容旧版
ai_codex_command、ai_claude_command、ai_codewhale_command配置,并迁移到结构化ai_cli选择。 - Codex 可启用
--ignore-user-config,避免桌面版config.toml中模型、推理强度或服务等级配置被不同版本 CLI 误解析。 - Codex 模型新增 GPT-5.6 Sol、Terra、Luna;默认使用
gpt-5.6-sol,默认推理强度为high。 - 推理强度按模型联动:Sol/Terra 支持到
ultra,Luna 支持到max,旧模型只展示其兼容选项。 - AI 日志会记录实际使用的 CLI 路径、版本、模型、推理强度、能力状态和候选拒绝原因,便于确认飞书任务的真实运行参数。
- 更新中英文 AI 配置文档,并补充跨平台扫描、严格选择、缓存失效、配置迁移和界面 API 测试。
- 应用内版本号更新为
v1.5.9。
安装包、便携包、Linux 包和 macOS 实验包会由 GitHub Actions 构建完成后自动上传到本 Release。
English
This release redesigns local AI Agent CLI discovery and selection, and adds Codex GPT-5.6 model support.
- Codex, Claude Code, and CodeWhale now share cross-platform CLI discovery, capability probing, and caching across Windows, macOS, Linux, and WSL.
- Automatic mode prefers the first compatible
PATHentry, then checks native installs, package-manager locations, and supported legacy paths. Detected and manual selections are strict and never silently fall back. - GUI, WebUI, and headless CLI surfaces now show candidate paths, versions, sources, compatibility states, testing, and explicit rescans.
- Added
ngawolf agent-cli list|test|rescan;ngawolf checkreports the exact resolved Agent CLI. - Legacy
ai_codex_command,ai_claude_command, andai_codewhale_commandsettings migrate into the structuredai_cliselection model. - Codex can use
--ignore-user-configso desktop changes to models, reasoning effort, or service tier do not break a differently versioned worker CLI. - Added GPT-5.6 Sol, Terra, and Luna. Codex now defaults to
gpt-5.6-solwithhighreasoning. - Reasoning choices follow model capabilities: Sol/Terra support through
ultra, Luna throughmax, and older models expose only compatible choices. - AI logs now record the actual CLI path, version, model, reasoning effort, capability state, and rejected candidates for reliable Feishu task diagnostics.
- Updated English and Chinese AI documentation and added coverage for cross-platform scanning, strict selection, cache invalidation, migration, and UI APIs.
- Updated the in-app version to
v1.5.9.
Installer, portable, Linux, and experimental macOS packages will be uploaded automatically by GitHub Actions after the release workflow finishes.