jihe520/agent-graph

选用指南

AI 根据 README 整理 · 2026/11/5 13:57:22

适用场景

适用用户：希望通过提示词让智能体改造自有应用交互界面的开发者。

查看 README 依据

- **Multiple harnesses on one screen**: Claude Code, Codex, Pi, and others appear on the same graph, told apart by color and avatar. See [DESIGN.md](DESIGN.md) for the full rules. ## Usage This assumes you can modify your ADE's source code, for example a fork of an open-source ADE like t3code, or your own harness frontend. Start an agent session in the ADE's source repository and send: ```text Clone https://github.com/jihe520/agent-graph into a temporary directory, read the PROMPT.md in it, and follow its instructions in the current project. ```

使用方式

使用方式：阅读 DESIGN.md、PROMPT.md 和参考原型 index.html，可将 PROMPT.md 交给智能体执行。

查看 README 依据

| [DESIGN.md](DESIGN.md) | Design spec: interaction model, visual encoding, rules for git operations, adaptation guidelines | | [PROMPT.md](PROMPT.md) | Implementation prompt: has an agent build this interaction model into your ADE, whatever the ADE or tech stack | | [index.html](index.html) | Reference prototype: a single file that demos every interaction with sample data; open it directly in a browser | DESIGN.md, PROMPT.md, and the prototype's UI are written in Chinese. Agents read them without trouble, and the prompt tells the agent to write UI text in your ADE's own language. ## Why

- **Multiple harnesses on one screen**: Claude Code, Codex, Pi, and others appear on the same graph, told apart by color and avatar. See [DESIGN.md](DESIGN.md) for the full rules. ## Usage This assumes you can modify your ADE's source code, for example a fork of an open-source ADE like t3code, or your own harness frontend. Start an agent session in the ADE's source repository and send: ```text Clone https://github.com/jihe520/agent-graph into a temporary directory, read the PROMPT.md in it, and follow its instructions in the current project. ```

部署要求

前提：需要能够修改目标应用源码，原型无需构建，可直接打开。

查看 README 依据

- **Multiple harnesses on one screen**: Claude Code, Codex, Pi, and others appear on the same graph, told apart by color and avatar. See [DESIGN.md](DESIGN.md) for the full rules. ## Usage This assumes you can modify your ADE's source code, for example a fork of an open-source ADE like t3code, or your own harness frontend. Start an agent session in the ADE's source repository and send: ```text Clone https://github.com/jihe520/agent-graph into a temporary directory, read the PROMPT.md in it, and follow its instructions in the current project. ```

- Run the prompt on your ADE, then open an issue with screenshots, your concept mapping table, and where it got stuck. That feedback drives revisions of DESIGN.md and PROMPT.md. - When you change an interaction, update the prototype and DESIGN.md together to keep them consistent. ## Modifying the prototype index.html has no dependencies and needs no build step: - Sample projects, branches, worktrees, and agents: `seed()`; sample conversations: `LOGS` - Harness names and colors: `HARNESS`; branch colors: `PALETTE`

已知限制

项目仍处于探索阶段，原型中数据均为模拟；明确不包含 git 客户端功能和会话 UI 重写。

查看 README 依据

- Still exploratory: both the design spec and the prompt will change. The next step is to get the prompt working end to end on t3code, then revise based on the results. - All data and live updates in the prototype are simulated. The 实时 (Live) and 重置 (Reset) controls in the top bar exist only for the prototype and are not part of the interaction model. - Out of scope: replacing your git client (viewing diffs, resolving conflicts) and rewriting the session UI (the host's existing one is reused). ## Contributing