refactor(*): 插件身份 - 将旧插件统一更名为 yingce
This commit is contained in:
ddCat
@@ -47,14 +47,14 @@ Canvas Agent 默认只监听 `127.0.0.1`。网页第一次带正确 token 连接
|
||||
|
||||
### Codex app 插件
|
||||
|
||||
仓库内提供了 Codex app 插件：`plugins/infinite-canvas`。该插件尚未上架公共插件目录，直接搜索不会显示；在 Codex app 中添加本仓库的 marketplace 后即可安装。插件会注册同一个 `infinite-canvas` MCP，并带上画布操作说明。
|
||||
仓库内提供了 Codex app 插件：`plugins/yingce`。该插件尚未上架公共插件目录，直接搜索不会显示；在 Codex app 中添加本仓库的 marketplace 后即可安装。插件会注册 `yingce` MCP，并带上画布操作说明。
|
||||
|
||||
添加本地 marketplace 时建议使用仓库绝对路径，避免 Codex 从其他工作目录解析失败：
|
||||
|
||||
```bash
|
||||
cd /path/to/infinite-canvas
|
||||
cd /path/to/open-ai-canvas
|
||||
codex plugin marketplace add "$(pwd)"
|
||||
codex plugin add infinite-canvas@infinite-canvas-local
|
||||
codex plugin add yingce@yingce-local
|
||||
```
|
||||
|
||||
插件默认通过 npm 启动 MCP：
|
||||
@@ -68,13 +68,13 @@ npx -y @ddcat666/open-ai-canvas-agent mcp
|
||||
Canvas Agent 启动后，给 Codex 添加 MCP：
|
||||
|
||||
```bash
|
||||
codex mcp add infinite-canvas -- npx -y @ddcat666/open-ai-canvas-agent mcp
|
||||
codex mcp add yingce -- npx -y @ddcat666/open-ai-canvas-agent mcp
|
||||
```
|
||||
|
||||
本仓库开发时可以改成，实际使用建议替换为本机绝对路径：
|
||||
|
||||
```bash
|
||||
codex mcp add infinite-canvas -- node /path/to/infinite-canvas/canvas-agent/dist/index.js mcp
|
||||
codex mcp add yingce -- node /path/to/open-ai-canvas/canvas-agent/dist/index.js mcp
|
||||
```
|
||||
|
||||
Canvas Agent 源码使用 TypeScript 编写，MCP 协议层使用官方 `@modelcontextprotocol/sdk`，工具入参使用 `zod` 描述。
|
||||
@@ -82,7 +82,7 @@ Canvas Agent 源码使用 TypeScript 编写，MCP 协议层使用官方 `@modelc
|
||||
如果希望终端里的 Codex 不被 MCP 审批卡住，可以在 `~/.codex/config.toml` 里给这个 MCP 设置自动放行：
|
||||
|
||||
```toml
|
||||
[mcp_servers.infinite-canvas]
|
||||
[mcp_servers.yingce]
|
||||
command = "npx"
|
||||
args = ["-y", "@ddcat666/open-ai-canvas-agent", "mcp"]
|
||||
default_tools_approval_mode = "approve"
|
||||
@@ -115,7 +115,7 @@ default_tools_approval_mode = "approve"
|
||||
|
||||
## 侧边栏 Codex
|
||||
|
||||
本地面板会把提示词发送给 Canvas Agent。Canvas Agent 使用官方 `@openai/codex` CLI 的 `codex app-server --stdio` 启动并复用同一个 Codex thread，启动时会注入 `infinite-canvas` MCP 配置并自动放行 MCP 审批，真正执行画布修改前仍由网页侧边栏二次确认。
|
||||
本地面板会把提示词发送给 Canvas Agent。Canvas Agent 使用官方 `@openai/codex` CLI 的 `codex app-server --stdio` 启动并复用同一个 Codex thread，启动时会注入 `yingce` MCP 配置并自动放行 MCP 审批，真正执行画布修改前仍由网页侧边栏二次确认。
|
||||
|
||||
侧边栏会展示 Codex 返回的 `thread.started`、`turn.started`、`item.*`、`turn.completed` 等结构化事件；收到 app-server 的 `item/agentMessage/delta` 时，Canvas Agent 会转成 `item.updated`，网页会用同一条消息做真实流式更新，并把工具细节收进运行日志。
|
||||
|
||||
@@ -128,13 +128,13 @@ Claude Code Adapter 代码暂时保留，但当前网页侧边栏只开放 Codex
|
||||
如果希望 Claude Code 也能操作画布，需要给 Claude Code 添加同一个 MCP。建议用 user scope，避免 Canvas Agent 从不同目录启动时找不到配置：
|
||||
|
||||
```bash
|
||||
claude mcp add --scope user --transport stdio infinite-canvas -- npx -y @ddcat666/open-ai-canvas-agent mcp
|
||||
claude mcp add --scope user --transport stdio yingce -- npx -y @ddcat666/open-ai-canvas-agent mcp
|
||||
```
|
||||
|
||||
本仓库开发时可以改成：
|
||||
|
||||
```bash
|
||||
claude mcp add --scope user --transport stdio infinite-canvas -- node /path/to/infinite-canvas/canvas-agent/dist/index.js mcp
|
||||
claude mcp add --scope user --transport stdio yingce -- node /path/to/open-ai-canvas/canvas-agent/dist/index.js mcp
|
||||
```
|
||||
|
||||
Canvas Agent 调用 Claude Code 时会默认带上 `--allowedTools mcp__infinite-canvas__*`，画布写操作仍由网页侧边栏确认。
|
||||
Canvas Agent 调用 Claude Code 时会默认带上 `--allowedTools mcp__yingce__*`，画布写操作仍由网页侧边栏确认。
|
||||
|
||||
@@ -95,7 +95,7 @@ export async function archiveCodexThread(emit: AgentEmit, threadId: string, cwd?
|
||||
|
||||
export function runClaudeTurn(prompt: string, emit: AgentEmit) {
|
||||
if (!prompt.trim()) return;
|
||||
const child = spawnAgent("claude", ["-p", "--output-format", "stream-json", "--verbose", "--include-partial-messages", "--allowedTools", "mcp__infinite-canvas__*", prompt], ["ignore", "pipe", "pipe"], emit);
|
||||
const child = spawnAgent("claude", ["-p", "--output-format", "stream-json", "--verbose", "--include-partial-messages", "--allowedTools", "mcp__yingce__*", prompt], ["ignore", "pipe", "pipe"], emit);
|
||||
if (!child) return;
|
||||
pipeJsonLines(child, emit, "claude");
|
||||
}
|
||||
@@ -277,7 +277,7 @@ export function canvasAgentMcpCommand() {
|
||||
export function codexConfig(configDir = CONFIG_DIR) {
|
||||
return {
|
||||
mcp_servers: {
|
||||
"infinite-canvas": {
|
||||
"yingce": {
|
||||
command: canvasAgentMcp.command,
|
||||
args: canvasAgentMcp.args,
|
||||
env: { FRAMEFIELD_LOCAL_RUNTIME_CONFIG_DIR: configDir },
|
||||
|
||||
@@ -11,7 +11,7 @@ export const DEFAULT_PORT = LOCAL_RUNTIME_DEFAULT_PORT;
|
||||
export const CONFIG_DIR = startupConfigDirectory();
|
||||
export const CONFIG_FILE = path.join(CONFIG_DIR, "canvas-agent.json");
|
||||
export const VERSION = readPackageVersion();
|
||||
export const AGENT_PROMPT = "你正在帮助用户操作影策网页画布。需要改动画布时优先使用已配置的 infinite-canvas MCP 工具：先 canvas_get_state 读取当前画布，再根据任务使用 canvas_create_text_node、canvas_generate_text、canvas_generate_image、canvas_generate_video、canvas_generate_audio、canvas_create_generation_flow、canvas_run_generation、canvas_update_node、canvas_connect_nodes 等通用工具；复杂批量改动再用 canvas_apply_ops，删除连线可用 delete_connections。需要生成内容时直接调用对应生成工具，不要绑定特定业务场景。即使用户明确点名 Dreamina/即梦，本机 Canvas Agent 也必须通过 canvas_generate_image 或 canvas_generate_video 进入共享 GenerationTask；本机图片模型使用 model=local:dreamina-cli:5.0 这类产品模型值，用户选择自动分辨率时使用 quality=auto；禁止调用 direct dreamina_cli provider tool。不要模拟鼠标点击，不要要求用户手动复制 JSON。";
|
||||
export const AGENT_PROMPT = "你正在帮助用户操作影策网页画布。需要改动画布时优先使用已配置的 yingce MCP 工具：先 canvas_get_state 读取当前画布，再根据任务使用 canvas_create_text_node、canvas_generate_text、canvas_generate_image、canvas_generate_video、canvas_generate_audio、canvas_create_generation_flow、canvas_run_generation、canvas_update_node、canvas_connect_nodes 等通用工具；复杂批量改动再用 canvas_apply_ops，删除连线可用 delete_connections。需要生成内容时直接调用对应生成工具，不要绑定特定业务场景。即使用户明确点名 Dreamina/即梦，本机 Canvas Agent 也必须通过 canvas_generate_image 或 canvas_generate_video 进入共享 GenerationTask；本机图片模型使用 model=local:dreamina-cli:5.0 这类产品模型值，用户选择自动分辨率时使用 quality=auto；禁止调用 direct dreamina_cli provider tool。不要模拟鼠标点击，不要要求用户手动复制 JSON。";
|
||||
|
||||
export type CanvasWorkspaceConfig = { workspacePath: string; activeThreadId?: string; pinnedThreadIds?: string[] };
|
||||
export type LocalRuntimeConfig = {
|
||||
|
||||
@@ -95,7 +95,7 @@ export function startLocalRuntime(options: StartLocalRuntimeOptions = {}) {
|
||||
await listening(server);
|
||||
log("Framefield Local Runtime");
|
||||
log("Runtime is listening on 127.0.0.1");
|
||||
log("Codex MCP: codex mcp add infinite-canvas -- npx -y @ddcat666/open-ai-canvas-agent mcp");
|
||||
log("Codex MCP: codex mcp add yingce -- npx -y @ddcat666/open-ai-canvas-agent mcp");
|
||||
} catch (startupError) {
|
||||
sessions.dispose();
|
||||
const cleanupErrors: unknown[] = [];
|
||||
|
||||
@@ -17,7 +17,7 @@ const config: CanvasAgentConfig = {
|
||||
type CommandFactory = () => { command: string; args: string[] };
|
||||
type CodexConfigFactory = (configDir?: string) => {
|
||||
mcp_servers: {
|
||||
"infinite-canvas": {
|
||||
"yingce": {
|
||||
command: string;
|
||||
args: string[];
|
||||
env?: Record<string, string>;
|
||||
@@ -44,7 +44,7 @@ test("internal Codex MCP config binds only the parent Runtime config directory",
|
||||
assert.equal(typeof configFactory, "function");
|
||||
if (!configFactory) return;
|
||||
const configDir = path.resolve("fixture-runtime-config-18871");
|
||||
const server = configFactory(configDir).mcp_servers["infinite-canvas"];
|
||||
const server = configFactory(configDir).mcp_servers["yingce"];
|
||||
assert.deepEqual(server.env, { FRAMEFIELD_LOCAL_RUNTIME_CONFIG_DIR: configDir });
|
||||
assert.equal(server.args.includes("--canvas-only"), true);
|
||||
assert.equal(server.args.some((value) => /token|secret/i.test(value)), false);
|
||||
@@ -55,7 +55,7 @@ test("internal MCP tool timeout covers the Canvas generation continuation window
|
||||
const configFactory = (agentsModule as unknown as { codexConfig?: CodexConfigFactory }).codexConfig;
|
||||
assert.equal(typeof configFactory, "function");
|
||||
if (!configFactory) return;
|
||||
const timeoutMs = configFactory().mcp_servers["infinite-canvas"].tool_timeout_sec * 1_000;
|
||||
const timeoutMs = configFactory().mcp_servers["yingce"].tool_timeout_sec * 1_000;
|
||||
assert.ok(timeoutMs >= 35 * 60 * 1_000, `internal MCP timeout ${timeoutMs}ms must cover the 35-minute generation continuation window`);
|
||||
assert.ok(timeoutMs > 35 * 60 * 1_000, "internal MCP timeout must leave shutdown/response margin after the Canvas continuation window");
|
||||
});
|
||||
|
||||
Reference in New Issue
Block a user