Agent助手

Name: Agent助手
Author: lightdash

AI 智能社区

作者仓库星标 5,841
作者更新于实时读取
作者仓库 lightdash

流狐档案事实层信息：领域、Agent、信任分、运行环境、权限与网络

领域: AI 智能
兼容 Agent: Claude Code
Cursor
Cline
Codex
Windsurf
Gemini CLI

+20
信任分: 88 / 100 · 社区维护
作者 / 版本 / 许可: @lightdash · 未声明 license
Token 消耗评级: 低消耗
接入复杂程度: 需手动接入
是否需要外部 API Key: 不需要
兼容的系统: Docker
底层运行要求: Node.js · Docker
文件与系统权限: 只读
允许写入 / 修改
Shell 执行
读取环境变量
网络行为: 仅限本地
安装命令数: 26 条

档案由构建时根据 SKILL.md 与安装命令自动衍生，可能与作者实际意图存在差异。

需要注意： 未限定 allowed-tools，默认拥有全部工具权限。

输出预览 agent-harness.preview

---
name: agent-harness
description: Guide for AI agents running in the isolated agent-harness environment. Use when you need to disc…
category: AI 智能
runtime: Node.js / Docker
---

# agent-harness 输出预览

## PART A: 任务判断
- 适用问题：提示词、Agent 工作流、模型评估或自动化推理。
- 输入要求：目标材料、限制条件、期望输出和验收方式。
- 证据边界：围绕“Step 1: Discover Your Agent ID / Step 2: Know Your Ports / Step 3: Use the CLI”读取原文规则，不把推断写成作者承诺。

## PART B: 执行结果
- **01** 任务判断：确认你的需求是否属于提示词、Agent 工作流、模型评估或自动化推理，并标出输入、限制和预期结果。
- **02** 执行计划：优先按“Step 1: Discover Your Agent ID / Step 2: Know Your Ports / Step 3: Use the CLI”拆成步骤，说明每一步会读取什么、修改什么、产出什么。
- **03** 交付结果：给出可复制的命令、文件改动、检查清单或内容草稿，并说明如何继续迭代。
- **04** 风险边界：结合 读取文件、写入/修改文件、执行终端命令、读取环境变量、主要在本地完成、通常不需要额外 API Key 给出执行前确认项。

## Running Rules
- 读取文件、写入/修改文件、执行终端命令、读取环境变量；主要在本地完成；通常不需要额外 API Key。
- 先小样例验证，再放大到真实任务。
- 交付时同时给结果、检查口径和下一步迭代建议。

SKILL.md SKILL.md

---
name: agent-harness
description: Guide for AI agents running in the isolated agent-harness environment. Use when you need to disc…
category: AI 智能
source: lightdash/lightdash
---

# agent-harness

## 什么时候使用
- 把 AI / Agent方向的常用动作沉淀成 Agent 可调用的技能 适合处理AI Agent、提示词、模型评估与自动化推理，核心价值是把输入、判断、执行、验证和交付边界固定下来，避免 Agent 泛泛回答。 把任务拆成可执行、可检查…
- 面向提示词、Agent 工作流、模型评估或自动化推理，优先处理能明确输入、步骤和验收标准的工作。

## 需要提供什么
- 目标材料、目录范围、期望结果和不可改动内容。
- 是否允许联网、执行命令、读写文件或调用外部服务。

## 执行规则
- 围绕「Step 1: Discover Your Agent ID / Step 2: Know Your Ports / Step 3: Use the CLI」组织步骤，不把推断写成作者事实。
- 读取文件、写入/修改文件、执行终端命令、读取环境变量；主要在本地完成；通常不需要额外 API Key。
- 先跑小样例，确认结果可检查后再扩大任务范围。

## 输出要求
- 给出最终产物、关键证据、验证方式和下一步动作。
- 信息不足时标记 unknown，不编造命令、平台或依赖。

证据边界与执行链路

作者原文负责流程事实；仓库文件负责来源和命令；流狐只补充适用场景、限制和质量判断。

skill "agent-harness" {
  输入层     -> 用户目标 + 目标文件 + 禁止范围 + 验收标准
  上下文层   -> Step 1: Discover Your Agent ID / Step 2: Know Your Ports / Step 3: Use the CLI
  规则层     -> SKILL.md 触发条件 / 执行顺序 / 输出格式
  运行层     -> Node.js / Docker | 读取文件、写入/修改文件、执行终端命令、读取环境变量 | 主要在本地完成
  安全层     -> 通常不需要额外 API Key + 小任务验证 + diff / 日志复核
  输出层     -> 可复制结果 + 检查清单 + 下一步迭代
}

解读按原文结构重写，命令、链接、术语均保留；右侧可核对作者原始 SKILL.md

先判断是否适合

核心用途：把 AI / Agent方向的常用动作沉淀成 Agent 可调用的技能适合处理AI Agent、提示词、模型评估与自动化推理，核心价值是把输入、判断、执行、验证和交付边界固定下来，避免 Agent 泛泛回答。把任务拆成可执行、可检查、可继续迭代的步骤…
适用任务：更适合 AI Agent、提示词、模型评估与自动化推理，尤其是输入、步骤、验收口径可以反复复用的场景。
不宜硬套：如果任务只有一次性问答、来源材料很少，或需要作者没有说明的命令和外部服务，应先降级为普通对话。

作者设计意图

触发依据：使用时机、上下文收集、输出格式和失败处理共同决定它是否适合当前任务。
来源边界：作者原文负责事实，仓库文件负责命令和文件证据；Fluxly 只补充用途、限制和使用判断。

作者的方法与取舍

执行主线：任务围绕「Step 1: Discover Your Agent ID / Step 2: Know Your Ports / Step 3: Use the CLI」组织，把触发条件、必要上下文、执行规则和交付标准连成一条可复核的链路。
设计重点：重点不在重复教程，而在解释作者为什么规定这些步骤、哪些条件不可省略，以及结果偏离时应回到哪一层检查。
使用判断：先确认任务与原文边界一致，再决定是否启用；来源没有说明的命令、平台和依赖保持未知，不由 Fluxly 补写。

边界和复核

依赖条件：通常不需要额外 API Key，适合先用小任务验证触发是否稳定。
权限提醒：会涉及 read / write / shell-exec / env-read 权限，执行前要说明可读写范围、命令边界和回退方式。
质量标准：结果应包含明确产物、必要证据和下一步动作；如果只是泛泛解释，需要补充输入和验收条件后重跑。

Agent Harness Quick Start

You are running in an isolated Lightdash development environment managed by the agent-harness. This guide helps you discover your environment and get productive immediately.

Step 1: Discover Your Agent ID

Your agent ID determines all your ports and resources. Find it by checking for the generated environment file:

ls -la .env.agent.* 2>/dev/null | head -1

This will show something like .env.agent.1 — the number is your agent ID.

Alternative: Check which PM2 processes are running:

pnpm exec pm2 jlist 2>/dev/null | node -e "
  const data = JSON.parse(require('fs').readFileSync(0, 'utf8'));
  const agents = [...new Set(data.map(p => p.name.match(/^agent-(\d+)-/)?.[1]).filter(Boolean))];
  if (agents.length === 1) console.log('AGENT_ID=' + agents[0]);
  else if (agents.length > 1) console.log('Multiple agents running: ' + agents.join(', '));
  else console.log('No agent processes found');
"

Step 2: Know Your Ports

Once you have your agent ID, your ports are deterministic:

Service	Formula	Agent 1	Agent 2	Agent 3
Frontend (Vite)	3000 + (ID × 10)	3010	3020	3030
Backend API	8000 + (ID × 10)	8010	8020	8030
Node Debugger	9200 + (ID × 10)	9210	9220	9230

Shared infrastructure (same for all agents):

PostgreSQL: localhost:15432
MinIO S3: localhost:19000
Headless Browser: localhost:13001
Mailpit SMTP: localhost:11025
Mailpit Web UI: localhost:18025

Get your URLs with:

./agent-harness/agent-cli.sh <AGENT_ID> url

Step 3: Use the CLI

All stack management goes through agent-cli.sh:

# Replace <ID> with your agent ID (1-5)

# Check process status (uptime, restarts, memory)
./agent-harness/agent-cli.sh <ID> status

# View logs (services: api, frontend, common-watch, warehouses-watch)
./agent-harness/agent-cli.sh <ID> logs api
./agent-harness/agent-cli.sh <ID> logs frontend

# Check API health
./agent-harness/agent-cli.sh <ID> health

# Restart services after code changes
./agent-harness/agent-cli.sh <ID> restart api
./agent-harness/agent-cli.sh <ID> restart frontend

# Run SQL queries against your database
./agent-harness/agent-cli.sh <ID> psql "SELECT * FROM users LIMIT 5;"

# View slow queries (useful for debugging)
./agent-harness/agent-cli.sh <ID> slow-queries

# Run arbitrary command with your agent's environment loaded
./agent-harness/agent-cli.sh <ID> exec pnpm -F backend test:dev:nowatch

Step 4: Verification Workflow

Run verification after every code change:

# Quick verification (typecheck + lint + unit tests for changed files)
./agent-harness/verify.sh <ID>

# Full verification (complete test suite + smoke test)
./agent-harness/verify.sh <ID> --full

Output is JSON to stdout:

{
  "status": "pass",
  "stages": [
    {"name": "typecheck", "status": "pass", "duration_ms": 3200},
    {"name": "lint", "status": "pass", "duration_ms": 1800},
    {"name": "test-unit", "status": "pass", "duration_ms": 4500}
  ],
  "total_duration_ms": 9500
}

Stages run in order, stopping on first failure:

typecheck — TypeScript for common, backend, frontend (parallel)
lint — ESLint for common, backend, frontend (parallel)
test-unit — Unit tests for changed files
test-full (--full only) — Complete test suite
smoke (--full only) — API health check

Email:    demo@lightdash.com
Password: demo_password!

Common Workflows

Making code changes

Edit files as needed
If you changed packages/common/, the TypeScript watcher auto-rebuilds
If you changed packages/backend/, restart the API:
```
./agent-harness/agent-cli.sh <ID> restart api
```
Run verification:
```
./agent-harness/verify.sh <ID>
```

Debugging API issues

# Check health
./agent-harness/agent-cli.sh <ID> health

# View recent API logs
./agent-harness/agent-cli.sh <ID> logs api

# Check for errors using Spotlight MCP
mcp__spotlight__search_errors with filters: {"timeWindow": 300}

# Get trace details
mcp__spotlight__get_traces with traceId: "<8-char-prefix-from-logs>"

Debugging frontend issues

# View Vite server logs
./agent-harness/agent-cli.sh <ID> logs frontend

# Use Chrome DevTools MCP for browser automation
# (Your frontend URL is http://localhost:<3000 + ID*10>)

Database queries

# View schema
./agent-harness/agent-cli.sh <ID> psql "\d tablename"

# Run queries
./agent-harness/agent-cli.sh <ID> psql "SELECT * FROM projects LIMIT 5;"

# Check slow queries
./agent-harness/agent-cli.sh <ID> slow-queries

Definition of Done

Your task is complete when ALL are true:

./agent-harness/verify.sh <ID> --full returns "status": "pass"
Feature/fix works in browser at your frontend URL
Tests cover the change
No TypeScript errors or lint warnings
Changes are committed

What NOT to Do

Do NOT modify shared infrastructure (PostgreSQL, MinIO containers)
Do NOT hardcode ports — always derive from your agent ID
Do NOT run docker compose commands — use agent-cli.sh
Do NOT touch other agents' databases or processes
Do NOT manually edit .env.agent.<ID> or ecosystem.agent.<ID>.config.cjs
Do NOT disable tests to make verification pass

Tmux Session Management

If you were launched with --claude, you're running inside a named tmux session agent-<ID>.

For human supervisors to attach to your session:

# Attach to a specific agent's tmux session
./agent-harness/attach.sh <ID>

# Or directly with tmux
tmux attach -t agent-<ID>

# List all agent sessions
tmux list-sessions | grep "^agent-"

Useful tmux commands (prefix is Ctrl-a on cloud servers, Ctrl-b locally):

prefix + d — Detach from session (leaves Claude running)
prefix + [ — Enter scroll mode (q to exit)
prefix + c — Create new window
prefix + n/p — Next/previous window

Quick Reference Card

# Status & Health
./agent-harness/agent-cli.sh <ID> status
./agent-harness/agent-cli.sh <ID> health
./agent-harness/agent-cli.sh <ID> url

# Logs
./agent-harness/agent-cli.sh <ID> logs api
./agent-harness/agent-cli.sh <ID> logs frontend

# Restart
./agent-harness/agent-cli.sh <ID> restart api
./agent-harness/agent-cli.sh <ID> restart frontend

# Database
./agent-harness/agent-cli.sh <ID> psql "<SQL>"
./agent-harness/agent-cli.sh <ID> slow-queries

# Verification
./agent-harness/verify.sh <ID>           # Quick
./agent-harness/verify.sh <ID> --full    # Complete

# Package commands (use your agent's env)
pnpm -F common typecheck
pnpm -F backend typecheck
pnpm -F frontend typecheck
pnpm -F backend lint
pnpm -F common test
pnpm generate-api  # After TSOA controller changes

安装命令

一键复制安装选择包管理器,点击右侧复制,粘贴到终端回车即可

npx skills add lightdash/lightdash/.claude/skills/agent-harness -y

查看通用安装教程

查看 Skill 原文

兼容工具

Claude Code
Cursor
Cline
Codex
Windsurf
Gemini
OpenCode
Roo Code
AAide
+Augment
GitHub Copilot
OpenClaw
KKiro
Junie
WWarp
Goose
Trae
Kilo Code
OpenHands
Amp
Zencoder
CContinue
AIAider
CodeBuddy
CodeGeeX
Replit Agent

安全审计

88 / 100

社区

未声明 license/许可。
未声明 version。
未限定 allowed-tools，默认拥有全部工具权限。

元信息

来源: github
lightdash/lightdash
English

分类与标签

分类 AI 智能