Skip to content

Edict for Mac 1.1.0

Choose a tag to compare

View all tags
@Logosss1 Logosss1 released this 24 Sep 18:34
· 42 commits to main since this release
v1.1.0
e4c4f14
This commit was signed with the committer’s verified signature.
claude Claude
SSH Key Fingerprint: 32dP45eSMmVSt/G/CGvcxl/P+MO3Nwj9xeTh/GSA2wc
Verified
Learn about vigilant mode.

Edict for Mac 1.1.0 · Apple Silicon (arm64)

What's new

  • Thinking-level slider in the edict box and the court: its steps are the selected model's own levels, and the far right is always that model's highest level. Set per model (OpenAI reasoning_effort / reasoning.effort, Claude output_config.effort or thinking.budget_tokens, Qwen, GLM, or custom JSON with levels like ultra) and per official. If a service rejects the parameter, it is dropped and the call retried once.
  • Readable model errors: relay errors such as HTTP 503 "this group doesn't support the model or access method" are no longer retried blindly. An error card explains the cause and offers retry with another model (re-runs only the failed step). Protocol probe in Models tests Chat / Responses / Messages with and without thinking parameters.
  • HTML preview: run generated pages in a sandboxed built-in browser with a console, device sizes, auto-reload and outbound-network blocking. Agents check their pages with the new preview_page tool (errors + screenshot).
  • Skills & MCP center: create, edit, enable and assign Skills; configure MCP servers with Cursor's mcp.json format (stdio, Streamable HTTP, legacy SSE), per-tool risk and approval, per-official grants. Local commands need your confirmation; secrets move into the macOS keychain.

Install: download Edict-1.1.0-arm64.dmg. Not signed with a Developer ID and not notarized — allow it once in System Settings → Privacy & Security → "Open Anyway". Your data folder is kept when you install over 1.0.0.

Verified: 41/41 unit + integration tests (mock LLM; MCP against a real process and real local HTTP servers), 20/20 + 19/19 end-to-end UI checks on Linux + Xvfb, DMG structure check. Not yet verified on a real Mac or with a real model service.

SHA-256 Edict-1.1.0-arm64.dmg: efca3d6610bdfc4c56b0da399549570671bd051433b5fc50239cbba562b36f23

中文说明

新功能

  • 思考程度滑块（工作台输入框与朝堂口谕框）：刻度就是所选模型自己的档位，最右 = 该模型最高档。可按模型设置（OpenAI、Claude、Qwen、GLM 或自定义 JSON，可加 ultra 等档位）与按官员设置；服务拒绝思考参数时自动去掉并重试一次。
  • 可读的模型错误：中转站 503「当前分组不支持该模型或接入方式」等配置错误不再盲目重试；错误卡片写明原因与做法，可「换模型重试」（只重跑失败的那一步）。「模型配置」新增「协议探测」。
  • HTML 预览：在沙箱内置浏览器中运行生成的网页：控制台、视口尺寸、自动刷新、默认拦截外网。官员用新工具 preview_page 截图自验。
  • 技能与 MCP 中心：技能增删改、启停、授权；MCP 使用与 Cursor 相同的 mcp.json 格式（stdio / Streamable HTTP / 旧版 SSE），逐工具风险与审批、按官员授权；本地命令须你确认，密钥自动存入钥匙串。

安装：下载 Edict-1.1.0-arm64.dmg。未签名、未公证，首次打开需在「系统设置 → 隐私与安全性」中点「仍要打开」。覆盖安装 1.0.0 会保留数据目录。

验证：单元 + 集成测试 41/41（mock LLM；MCP 使用真实子进程与本地 HTTP 服务），端到端 UI 20/20 + 19/19（Linux + Xvfb），DMG 结构校验通过。尚未在真实 Mac 与真实模型服务上验证。

Assets 4
Loading