feat: 新增 Hook 遥测回调 + 更新 README

scripts/run.ts 在每次 skill 执行后自动 POST hookUrl(由 auth server 下发),
实现服务端可观测性,客户端零配置。README 补充遥测机制、新建 skill 流程说明。

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
This commit is contained in:
ywkj 2026-04-20 07:39:46 +08:00
parent 8be28aab5a
commit 01fcb90505
2 changed files with 152 additions and 36 deletions

128
README.md
View File

@ -1,20 +1,19 @@
# template-skill # template-skill
新 skill 的基础模版。 新 skill 的基础模版内置认证、遥测回调、CI/CD 注册流程。
---
## 认证机制auth-cli.ts ## 认证机制auth-cli.ts
每个 skill 内置一份 `src/auth-cli.ts`,它是一个薄 wrapper通过 subprocess 调用 `auth-rt` 二进制。 每个 skill 内置一份 `src/auth-cli.ts`,通过 subprocess 调用 `auth-rt` 二进制完成认证,**零 npm 依赖**。
**不使用 npm 依赖**auth-runtime 更新时只需重新编译二进制,不需要改动任何 skill。
### 工作原理
``` ```
skill/src/index.ts skill/scripts/run.ts
→ import { createSkillClient } from './auth-cli.ts' → createSkillClient().session()
→ auth-cli.ts 通过 spawnSync 调用 auth-rt 二进制 → auth-cli.ts spawnSync auth-rt
→ auth-rt 处理 token/session/request → auth-rt 读取 ~/.openclaw/.env (CLIENT_KEY)
→ 返回 { accessToken, hookUrl, hookToken, ... }
``` ```
### 使用方式 ### 使用方式
@ -22,44 +21,105 @@ skill/src/index.ts
```typescript ```typescript
import { createSkillClient } from './auth-cli.ts'; import { createSkillClient } from './auth-cli.ts';
const client = createSkillClient({ const client = createSkillClient({ dryRun: false });
apiBase: process.env.API_BASE, // 可选 const session = await client.session();
dryRun: false, // 可选dry-run 模式返回模拟数据 // session = { accessToken, hookUrl, hookToken, ... }
});
// API 调用
const res = await client.post('/ecom/your/endpoint', { param: 'value' }); const res = await client.post('/ecom/your/endpoint', { param: 'value' });
// res = { status: 200, body: '...' } // res = { status: 200, body: '...' }
// 获取 session
const session = await client.session();
// session = { accessToken: '...', expiresIn: 900 }
``` ```
### 前置条件 ### 前置条件
每台运行 skill 的机器上必须安装 `auth-rt` 二进制 每台运行 skill 的机器需安装 `auth-rt``install.sh` 自动处理)
```bash ```bash
git clone http://192.168.0.108:3030/agent-skills/auth-runtime.git ~/clawd/skills/auth-runtime # 手动安装
cd ~/clawd/skills/auth-runtime && ./install.sh bash install.sh
# 安装到 ~/.openclaw/bin/auth-rt
# 或确保 ~/.openclaw/.env 中配置
CLIENT_KEY=sk_xxxx
``` ```
确保 `~/.openclaw/bin` 在 PATH 中,或通过 `AUTH_RT_BIN` 环境变量指定路径。 ---
### auth-runtime 更新流程 ## 遥测与可观测性Hook 回调
`scripts/run.ts` 在每次 skill 执行后,**自动**向 auth server 下发的 `hookUrl` 发送结构化执行报告。
### 工作原理
```
Client 执行 skill
├─ 1. auth-rt session → auth gateway 返回 { hookUrl, hookToken }
├─ 2. 执行命令
└─ 3. POST hookUrl (fire-and-forget不阻塞输出)
{
skill: "my-skill",
command: "run",
status: "success" | "failed",
durationMs: 1234,
error: "..." (仅失败时)
}
服务端 → Loki → Grafana
```
### 特性
- **服务端控制**`hookUrl` 由 auth server 在 session 时下发,客户端无需任何配置
- **非阻塞**fire-and-forgethook 失败不影响 skill 正常执行和输出
- **dry-run 跳过**`--dry-run` 模式下不发送 hook
- **统一入口**:所有遥测逻辑在 `scripts/run.ts` 中,`src/index.ts` 保持纯业务逻辑
### 服务端可查询内容
| 维度 | Loki label / 字段 |
|------|-----------------|
| 哪个客户端调用了哪个 skill | `CLIENT_KEY` → session → hook |
| 成功率 / 失败率 | `status` |
| 响应耗时 | `durationMs` |
| 错误类型分布 | `error` |
| 命令维度拆分 | `command` |
---
## 新建 Skill 检查清单
1. 从此模版创建仓库Forgejo → Use this template
2. `package.json` 中修改 `name` 字段
3. `scripts/run.ts` 中修改 `SKILL_NAME` 常量
4. `SKILL.md` 中更新 `name` / `description` frontmatter
5. `src/index.ts` 中实现业务逻辑
6. **不要**`package.json` 中添加 `@clawd/auth-runtime` 依赖
7. 推送 `v*` tag 触发 CI/CD 自动注册 skill
---
## CI/CD自动注册
`.forgejo/workflows/register-skill-release.yml` 在推送 `v*` tag 时触发,调用 `shared-actions/register-skill` 将 skill 注册到平台。
auth-runtime 代码变更后:
```bash ```bash
cd ~/clawd/skills/auth-runtime && git pull && ./install.sh # 发布新版本
git tag v1.0.0 && git push origin v1.0.0
``` ```
重新编译即可,**无需改动任何 skill 代码**。
### 新建 skill 检查清单 注册所需的 `CLIENT_KEY` 通过 Forgejo 仓库 Secret 配置skill 代码中不包含任何密钥。
1. 从此模版创建仓库 ---
2. 确认 `src/auth-cli.ts` 已包含(直接从模版继承)
3. `src/index.ts``import { createSkillClient } from './auth-cli.ts'` ## 目录结构
4. `package.json`**不要** 添加 `@clawd/auth-runtime` 依赖
5. `install.sh` 中包含 auth-rt 二进制检查 ```
.forgejo/workflows/ # CI/CDtag 触发自动注册
scripts/run.ts # CLI 入口:参数解析 + session + 遥测 hook
src/
auth-cli.ts # auth-rt 薄 wrapper勿修改
index.ts # 业务逻辑入口
types.ts # 类型定义
SKILL.md # skill 元数据frontmatter+ Claude 使用说明
.env.example # 环境变量模版
install.sh # 依赖安装(含 auth-rt 自动下载)
```

View File

@ -1,6 +1,9 @@
#!/usr/bin/env bun #!/usr/bin/env bun
import type { Command } from '../src/index.ts'; import type { Command } from '../src/index.ts';
import { run } from '../src/index.ts'; import { run } from '../src/index.ts';
import { createSkillClient } from '../src/auth-cli.ts';
const SKILL_NAME = 'my-skill'; // TODO: replace with actual skill name
function printUsage(): void { function printUsage(): void {
console.error(`Usage: console.error(`Usage:
@ -9,10 +12,20 @@ function printUsage(): void {
Commands: Commands:
run <arg> run <arg>
Config: ~/.openclaw/.env (API_BASE) Config: ~/.openclaw/.env (CLIENT_KEY)
`); `);
} }
async function reportHook(
hookUrl: string,
hookToken: string | undefined,
payload: object,
): Promise<void> {
const headers: Record<string, string> = { 'Content-Type': 'application/json' };
if (hookToken) headers['Authorization'] = `Bearer ${hookToken}`;
await fetch(hookUrl, { method: 'POST', headers, body: JSON.stringify(payload) });
}
async function main(): Promise<void> { async function main(): Promise<void> {
const positionals: string[] = []; const positionals: string[] = [];
let dryRun = false; let dryRun = false;
@ -31,8 +44,51 @@ async function main(): Promise<void> {
if (positionals.length < 1) { printUsage(); process.exit(1); } if (positionals.length < 1) { printUsage(); process.exit(1); }
const result = await run(positionals[0] as Command, positionals.slice(1), dryRun); const command = positionals[0] as Command;
// Exchange CLIENT_KEY for session — gives us hookUrl for telemetry
let hookUrl: string | undefined;
let hookToken: string | undefined;
try {
const client = createSkillClient({ dryRun });
const session = await client.session();
hookUrl = session.hookUrl;
hookToken = session.hookToken;
} catch {
// Auth failure is non-fatal for telemetry; skill still runs
}
const startMs = Date.now();
let result: Awaited<ReturnType<typeof run>>;
try {
result = await run(command, positionals.slice(1), dryRun);
} catch (err) {
const error = err instanceof Error ? err.message : String(err);
const failed = { status: 'failed' as const, command, dryRun, error };
console.log(JSON.stringify(failed, null, 2));
if (hookUrl && !dryRun) {
reportHook(hookUrl, hookToken, {
skill: SKILL_NAME, command, status: 'failed',
durationMs: Date.now() - startMs, error,
}).catch(() => {});
}
process.exit(1);
}
console.log(JSON.stringify(result, null, 2)); console.log(JSON.stringify(result, null, 2));
// Fire-and-forget telemetry — never delays output
if (hookUrl && !dryRun) {
reportHook(hookUrl, hookToken, {
skill: SKILL_NAME,
command,
status: result.status,
durationMs: Date.now() - startMs,
error: (result as any).error,
}).catch(() => {});
}
} }
main().catch((err) => { main().catch((err) => {