ClaudeMods
☰
ZH-TW
● 0 人在線上 · 瀏覽 0 次
贊助提交作品
GitHub 儲存庫 · 發布者 ThomasRohde

nendo-planner

顯示哪些 .nendo 檔案已開啟的狀態列,以及顯示規劃器 Now lane 的提示列。

ThomasRohde@ThomasRohde

ThomasRohde/nendo/tree/main/.claude/skills/nendo-planner

已翻譯

關於這個 mod

顯示哪些 .nendo 檔案已開啟的狀態列,以及顯示規劃器 Now lane 的提示列

安裝

請先查看作者 README,確認 marketplace 與外掛名稱;指令可能隨儲存庫結構而變動。

claude plugin marketplace add ThomasRohde/nendo
claude plugin install nendo-planner
原文 / README
<p align="center"> <img src="docs/assets/brand/nendo.png" alt="Nendo" width="420"> </p>

Nendo

Malleable software in a single file. One portable SQLite file starts empty and is shaped — by people and by coding agents — into a working application: schema, data, forms, boards, record pages and revision history. No generated project. No build step. No regeneration when you want a change.

Nendo (粘土) is Japanese for clay.

Experimental. Nendo is a research prototype exploring whether malleable software works, not a product. It runs as an unsigned per-user Windows x64 install. Read State before you rely on anything here.

thomasrohde.github.io/nendo — the concept, how it works, how to use it, the honest status, and this repository's documentation rendered as a site.

<p align="center"> <img src="docs/reviews/2026-09-12-mcp-vocabulary/production-board-dark.png" alt="A Nendo board surface grouped by a choice field, with column totals, in the dark theme" width="860"> </p> <p align="center"><sub>A board an agent authored over MCP, in the running host. Captured from the 2026-09-12 review build (0.17.0 is current), with the rest of that review in <a href="docs/reviews/README.md">docs/reviews/</a>.</sub></p>

The installed host always provides Nendo Studio, a high-quality database explorer and editor. Custom surfaces add focused experiences on top, but never become the only route to the data — if a surface breaks, the data is still there, and still editable.

How it works

Create empty file
  → add or import schema and data
  → work in the default Studio table
  → ask an agent to create or reshape a form, board or record page
  → review the semantic diff
  → accept, observe, and compensate where the operation supports it

Agents connect over a local MCP interface. They never get SQL, a database path or filesystem access — only typed semantic operations. Their changes are validated on a physical clone and previewed as a readable diff; accepting replays the exact validated operations against your file. Nothing an agent gets wrong can reach your data before you say yes.

Start here

  • Vision — what malleable software means and what would falsify it
  • Architecture — the system as built
  • Roadmap — what's next, and what is honestly not yet true
  • Development planner — how this project plans its own work in a Nendo file. That file is the author's own data and is not in a clone; the document describes the workflow and the agent handoff
  • Decisions — the ADRs, which are the architecture authority
  • Contracts — behavioural detail: MCP interface, semantic surfaces, scalars, queries, CSV, relationships, reads and authority
  • Glossary — the vocabulary this repository uses precisely

State

The MVP loop works end to end, delivered as a local unsigned per-user Windows x64 install. Four reference applications falsify the hypothesis from different shapes — Idea Garden, Decision Log, the Axiom Register and Nendo Station — and the last two were built from an empty file through the MCP interface alone.

Public distribution, signing, ARM64, cloud sync and cross-platform support are not qualified, and the independent human evaluation has not been run. The roadmap states each gap plainly.

Build

Windows x64 only. Nendo.Engine is net10.0 and builds anywhere, but the Desktop host is WinUI 3 with WebView2.

Prerequisites

| Tool | Version | Install | | --- | --- | --- | | .NET SDK | Pinned in global.json | winget install Microsoft.DotNet.SDK.10 | | Node.js with npm | The engines range in src/Nendo.Workbench/package.json | winget install OpenJS.NodeJS.LTS | | PowerShell 7 | pwsh; every script in tools/ assumes it | winget install Microsoft.PowerShell | | Git | Any current version | winget install Git.Git | | WebView2 Evergreen runtime | Already present on Windows 11 | winget install Microsoft.EdgeWebView2Runtime | | NSIS | Only to build the installer; makensis on PATH | winget install NSIS.NSIS, or the NuGet route below |

Open a new shell after installing so PATH picks the tools up.

Where Group Policy disables winget, or only an internal package mirror is an approved source, NSIS is also published on NuGet as the portable package NSIS-Tool (a third-party repackaging; its makensis.exe is not Authenticode-signed, the package carries NuGet.org's repository signature). Restore it through your mirror and put it on PATH for the build session only; nothing is installed. The steps are in architecture.md. The Windows App SDK is bundled into the build output (self-contained), so it needs no separate runtime. global.json rolls forward to the latest feature band, so a newer 10.0 SDK that is already installed wins the pin.

Build and run

cd src/Nendo.Workbench; npm ci; npm run build; cd ../..
dotnet build Nendo.slnx
./artifacts/bin/Nendo.Desktop/debug_win-x64/Nendo.Desktop.exe

Do not skip the Workbench step. The Desktop project copies src/Nendo.Workbench/dist into its output but does not build it, so dotnet build on a fresh clone succeeds and produces a host with no interface to show. All build output goes under artifacts/, which is git-ignored.

Started with no argument, the app opens without a file; create one from there. Pass the path of a .nendo file to open it instead. The files in workspace/ are tracked demos: open a copy, not the original.

Verify

pwsh ./tools/Test-Repository.ps1    # fast invariant check
pwsh ./tools/Test-Production.ps1    # full gate; includes the above

Test-Production.ps1 does the whole build itself (npm ci, Workbench type check, tests and build, then .NET restore, build and tests), so on a fresh clone it is also the one-command build. It takes about a minute once packages are restored. Some Desktop tests open real windows and use the clipboard, so run the gate in an unlocked, interactive desktop session. On a locked workstation or a headless session those tests fail without any product defect.

Packaging, the installer and the native review lanes outside the gate are documented in architecture.md.

Connect an agent

While a file is open with Agent access on, Nendo listens on a loopback port that the file keeps on this computer. The first file you switch access on for keeps 41763, and each further file keeps the next free port. The address is the whole client configuration — there is no credential to find or paste:

claude mcp add --transport http nendo http://127.0.0.1:41763/mcp
codex mcp add nendo --url http://127.0.0.1:41763/mcp

Agent → Connection shows the address a file uses and copies either command.

The registrations checked into this repository (.mcp.json and .codex/config.toml) point at port 41766. That is the port the author's planner file keeps on their machine, not a default, so on a fresh machine they connect to nothing. To use them, set Port for this file to 41766 in Agent → Connection for the file you want an agent in this checkout to reach; otherwise register the address that panel shows.

Anything running on this computer can connect at the chosen access level, so leave access Off when no agent is working.

For coding agents

Codex reads AGENTS.md; CLAUDE.md imports that same file for Claude Code. Both use Nendo Development as the primary work planner through the repository's existing MCP registrations. Read the live work item and its acceptance criteria before implementation, and record outcomes and remaining work at handoff. Accepted ADRs retain architecture authority.

The planner file is the author's own data and is not in a clone. Without it there is no live work item to read: follow the planner-unavailable rule in AGENTS.md, work from the repository instructions and the task in hand, and do not create a replacement planner.

A curated, pinned set of first-party .NET agent skills is vendored under .agents/skills/ — use the matching skill rather than guessing current SDK, template, MSBuild or test behaviour.

Repository layout

| Path | What is in it | | --- | --- | | src/Nendo.Engine | The typed core: storage, semantic operations, diff, behaviour. The only code that touches SQLite | | src/Nendo.Desktop | The WinUI 3 host: window, native shell integration, serving custom views from the open file | | src/Nendo.Workbench | The renderer — TypeScript and Vite, bundled into the Desktop output | | src/Nendo.LocalMcp | The loopback MCP server: tools, resources, leases and access modes | | tests/ | MSTest suites for the Engine, the Desktop host and the MCP adapter | | tools/ | Build, packaging, gate and review scripts (PowerShell and Node) | | extensions/ | Source for the four example custom-view packages; a file carries a package's code once it is imported | | fixtures/ | Reference-application seed data | | workspace/ | Tracked .nendo demo files | | docs/ | Vision, architecture, ADRs, contracts and reviews | | site/ | The public website (ADR-0018) — Astro, outside the product boundary |

Contributing and security

CONTRIBUTING.md has the loop, the gate to run and the house rules. SECURITY.md states the trust boundary and how to report a vulnerability.

Brand

Static mark · Animated mark

Licence

MIT. Vendored third-party material retains its own notices and licences.

更多類似作品