AGENTS.md vs SOUL.md 区别

特性	AGENTS.md	SOUL.md
作用	工作空间规则和操作指南	Agent 的身份、性格和核心行为准则
内容	任务流程、通信协议、安全规范、工具使用说明	角色定义、价值观、行为风格、自我认知
谁制定	系统/Manager 生成 + 用户自定义规则	用户为每个 Agent 量身定制
层级	组织/工作空间级别	个人/Agent 级别
修改权限	可追加自定义规则	完全可定制
目的	告诉 Agent 如何工作	告诉 Agent 它是谁

具体对比

AGENTS.md - "工作手册"

解决什么问题：

• 怎么接收任务？
• 怎么和 Manager/其他 Worker 沟通？
• 文件放哪里？
• 什么时候 @mention？
• 安全规范是什么？

SOUL.md - "灵魂/人格定义"

解决什么问题：

• 我是谁？
• 我的专业领域是什么？
• 我该怎么说话？（语气、风格）
• 我的价值观是什么？
• 面对不同情况该怎么反应？

实际例子

paulwong 2026-04-16 16:26 发表评论

paulwong 2026-04-15 11:08 发表评论

name: firecrawl

x-common-service: &common-service
  # NOTE: If you don't want to build the service locally,
  # comment out the build: statement and uncomment the image: statement
  image: ghcr.io/firecrawl/firecrawl:latest
  # build: apps/api

  ulimits:
    nofile:
      soft: 65535
      hard: 65535
  networks:
    - backend
  extra_hosts:
    - "host.docker.internal:host-gateway"
  logging:
    driver: "json-file"
    options:
      max-size: "10m"
      max-file: "3"
      compress: "true"

x-common-env: &common-env
  REDIS_URL: ${REDIS_URL:-redis://redis:6379}
  REDIS_RATE_LIMIT_URL: ${REDIS_URL:-redis://redis:6379}
  PLAYWRIGHT_MICROSERVICE_URL: ${PLAYWRIGHT_MICROSERVICE_URL:-http://playwright-service:3000/scrape}
  POSTGRES_USER: ${POSTGRES_USER:-postgres}
  POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:-postgres}
  POSTGRES_DB: ${POSTGRES_DB:-postgres}
  POSTGRES_HOST: ${POSTGRES_HOST:-nuq-postgres}
  POSTGRES_PORT: ${POSTGRES_PORT:-5432}
  USE_DB_AUTHENTICATION: ${USE_DB_AUTHENTICATION:-false}
  NUM_WORKERS_PER_QUEUE: ${NUM_WORKERS_PER_QUEUE:-8}
  CRAWL_CONCURRENT_REQUESTS: ${CRAWL_CONCURRENT_REQUESTS:-10}
  MAX_CONCURRENT_JOBS: ${MAX_CONCURRENT_JOBS:-5}
  BROWSER_POOL_SIZE: ${BROWSER_POOL_SIZE:-5}
  OPENAI_API_KEY: ${OPENAI_API_KEY}
  OPENAI_BASE_URL: ${OPENAI_BASE_URL}
  MODEL_NAME: ${MODEL_NAME}
  MODEL_EMBEDDING_NAME: ${MODEL_EMBEDDING_NAME}
  OLLAMA_BASE_URL: ${OLLAMA_BASE_URL}
  AUTUMN_SECRET_KEY: ${AUTUMN_SECRET_KEY}
  SLACK_WEBHOOK_URL: ${SLACK_WEBHOOK_URL}
  BULL_AUTH_KEY: ${BULL_AUTH_KEY}
  TEST_API_KEY: ${TEST_API_KEY}
  SUPABASE_ANON_TOKEN: ${SUPABASE_ANON_TOKEN}
  SUPABASE_URL: ${SUPABASE_URL}
  SUPABASE_SERVICE_TOKEN: ${SUPABASE_SERVICE_TOKEN}
  SELF_HOSTED_WEBHOOK_URL: ${SELF_HOSTED_WEBHOOK_URL}
  LOGGING_LEVEL: ${LOGGING_LEVEL}
  PROXY_SERVER: ${PROXY_SERVER}
  PROXY_USERNAME: ${PROXY_USERNAME}
  PROXY_PASSWORD: ${PROXY_PASSWORD}
  SEARXNG_ENDPOINT: ${SEARXNG_ENDPOINT}
  SEARXNG_ENGINES: ${SEARXNG_ENGINES}
  SEARXNG_CATEGORIES: ${SEARXNG_CATEGORIES}
  MAX_CPU: ${MAX_CPU:-0.8}
  MAX_RAM: ${MAX_RAM:-0.8}
  ALLOW_LOCAL_WEBHOOKS: ${ALLOW_LOCAL_WEBHOOKS:-false}

services:
  playwright-service:
    # NOTE: If you don't want to build the service locally,
    # comment out the build: statement and uncomment the image: statement
    image: ghcr.io/firecrawl/playwright-service:latest
    # build: apps/playwright-service-ts
    environment:
      PORT: 3000
      PROXY_SERVER: ${PROXY_SERVER}
      PROXY_USERNAME: ${PROXY_USERNAME}
      PROXY_PASSWORD: ${PROXY_PASSWORD}
      ALLOW_LOCAL_WEBHOOKS: ${ALLOW_LOCAL_WEBHOOKS:-false}
      BLOCK_MEDIA: ${BLOCK_MEDIA:-false}
      # Configure maximum concurrent pages for Playwright browser instances
      MAX_CONCURRENT_PAGES: ${CRAWL_CONCURRENT_REQUESTS:-10}
    networks:
      - backend
    # Resource limits for Docker Compose (not Swarm)
    cpus: 2.0
    mem_limit: 4G
    memswap_limit: 4G
    logging:
      driver: "json-file"
      options:
        max-size: "10m"
        max-file: "3"
        compress: "true"
    tmpfs:
      - /tmp/.cache:noexec,nosuid,size=1g

  api:
    <<: *common-service
    environment:
      <<: *common-env
      HOST: "0.0.0.0"
      PORT: ${INTERNAL_PORT:-3002}
      EXTRACT_WORKER_PORT: ${EXTRACT_WORKER_PORT:-3004}
      WORKER_PORT: ${WORKER_PORT:-3005}
      NUQ_RABBITMQ_URL: amqp://rabbitmq:5672
      ENV: local
    depends_on:
      redis:
        condition: service_started
      playwright-service:
        condition: service_started
      rabbitmq:
        condition: service_healthy
      nuq-postgres:
        condition: service_started
    ports:
      - "${PORT:-3002}:${INTERNAL_PORT:-3002}"
    command: node dist/src/harness.js --start-docker
    # Resource limits for Docker Compose (not Swarm)
    # Increase if you have more CPU cores/RAM available
    cpus: 4.0
    mem_limit: 8G
    memswap_limit: 8G

  redis:
    # NOTE: If you want to use Valkey (open source) instead of Redis (source available),
    # uncomment the Valkey statement and comment out the Redis statement.
    # Using Valkey with Firecrawl is untested and not guaranteed to work. Use with caution.
    image: redis:alpine
    # image: valkey/valkey:alpine

    networks:
      - backend
    command: redis-server --bind 0.0.0.0
    volumes:
      - redis-data:/data
    logging:
      driver: "json-file"
      options:
        max-size: "5m"
        max-file: "2"
        compress: "true"

  rabbitmq:
    image: rabbitmq:3-management
    networks:
      - backend
    command: rabbitmq-server
    healthcheck:
      test: ["CMD", "rabbitmq-diagnostics", "-q", "check_running"]
      interval: 10s
      timeout: 10s
      retries: 10
      start_period: 30s
    volumes:
      - rabbitmq-data:/var/lib/rabbitmq
    logging:
      driver: "json-file"
      options:
        max-size: "5m"
        max-file: "2"
        compress: "true"

  nuq-postgres:
    image: ghcr.io/firecrawl/nuq-postgres:latest
    environment:
      POSTGRES_USER: ${POSTGRES_USER:-postgres}
      POSTGRES_PASSWORD: ${POSTGRES_PASSWORD:-postgres}
      POSTGRES_DB: ${POSTGRES_DB:-postgres}
    networks:
      - backend
    volumes:
      - postgres-data:/var/lib/postgresql/data
    logging:
      driver: "json-file"
      options:
        max-size: "10m"
        max-file: "3"
        compress: "true"

networks:
  backend:
    driver: bridge

volumes:
  redis-data:
    driver: local
  rabbitmq-data:
    driver: local
  postgres-data:
    driver: local

# ===== Required ENVS ======
PORT=8019
HOST=0.0.0.0

# Note: PORT is used by both the main API server and worker liveness check endpoint

# To turn on DB authentication, you need to set up Supabase.
USE_DB_AUTHENTICATION=false

# ===== Optional ENVS ======

## === AI features (JSON format on scrape, /extract API) ===
# Provide your OpenAI API key here to enable AI features
# OPENAI_API_KEY=your_openai_api_key_here

# Experimental: Use Ollama
# OLLAMA_BASE_URL=http://localhost:11434/api
# MODEL_NAME=deepseek-r1:7b
# MODEL_EMBEDDING_NAME=nomic-embed-text

# Experimental: Use any OpenAI-compatible API
# OPENAI_BASE_URL=https://example.com/v1
# OPENAI_API_KEY=your_api_key_here

## === Proxy ===
# PROXY_SERVER can be a full URL (e.g. http://0.1.2.3:1234) or just an IP and port combo (e.g. 0.1.2.3:1234)
# Do not uncomment PROXY_USERNAME and PROXY_PASSWORD if your proxy is unauthenticated
# PROXY_SERVER=
# PROXY_USERNAME=
# PROXY_PASSWORD=

## === /search API ===
# By default, the /search API will use Google search.

# You can specify a SearXNG server with the JSON format enabled, if you'd like to use that instead of direct Google.
# You can also customize the engines and categories parameters, but the defaults should also work just fine.
# SEARXNG_ENDPOINT=http://your.searxng.server
# SEARXNG_ENGINES=
# SEARXNG_CATEGORIES=

## === PostgreSQL Database Configuration ===
# Configure PostgreSQL credentials. These should match the credentials used by the nuq-postgres container.
# If you change these, ensure all three are set consistently.
# Note: nuq-postgres requires using 'postgres' as the database name for proper pg_cron initialization
POSTGRES_USER=postgres
POSTGRES_PASSWORD=postgres
POSTGRES_DB=postgres

## === Redis Configuration ===
# These are auto-configured by docker-compose.yaml. You shouldn't need to change them.
# REDIS_URL=redis://redis:6379
# REDIS_RATE_LIMIT_URL=redis://redis:6379

## === Playwright Service ===
# This is auto-configured by docker-compose.yaml. You shouldn't need to change it.
# PLAYWRIGHT_MICROSERVICE_URL=http://playwright-service:3000/scrape

## === Supabase Setup (used to support DB authentication, advanced logging, etc.) ===
# SUPABASE_ANON_TOKEN=
# SUPABASE_URL=
# SUPABASE_SERVICE_TOKEN=

# Use if you've set up authentication and want to test with a real API key
# TEST_API_KEY=

# This key lets you access the queue admin panel. Change this if your deployment is publicly accessible.
BULL_AUTH_KEY=CHANGEME

## === PDF Parsing ===
# Set if you have a llamaparse key you'd like to use to parse pdfs
# LLAMAPARSE_API_KEY=

## === Monitoring ===
# Set if you'd like to send server health status messages to Slack
# SLACK_WEBHOOK_URL=

# Set if you'd like to send posthog events like job logs
# POSTHOG_API_KEY=
# POSTHOG_HOST=

## === System Resource Configuration ===
# Maximum CPU usage threshold (0.0-1.0). Worker will reject new jobs when CPU usage exceeds this value.
# Default: 0.8 (80%)
MAX_CPU=0.8

# Maximum RAM usage threshold (0.0-1.0). Worker will reject new jobs when memory usage exceeds this value.
# Default: 0.8 (80%)
MAX_RAM=0.8

# Number of workers per queue
NUM_WORKERS_PER_QUEUE=8

# Concurrent requests for crawling
CRAWL_CONCURRENT_REQUESTS=10

# Maximum concurrent jobs
MAX_CONCURRENT_JOBS=5

# Browser pool size
BROWSER_POOL_SIZE=5

# Set if you'd like to allow local webhooks to be sent to your self-hosted instance
# ALLOW_LOCAL_WEBHOOKS=true

# Block media in Playwright
# BLOCK_MEDIA=true

# Logging level (DEBUG, INFO, WARN, ERROR)
LOGGING_LEVEL=INFO

paulwong 2026-04-15 11:03 发表评论

https://skillhub.cn/skills/llm-wiki

clawhub上的

https://clawhub.ai/john-ver/karpathy-llm-wiki

深加工成知识图谱的graphify

https://github.com/safishamsi/graphify/blob/v4/README.zh-CN.md

微信公众号文章拉取技能

https://skillhub.cn/skills/wechat-article-spider

将html转成markdown的技能

https://skillhub.cn/skills/markdown-converter

paulwong 2026-04-12 17:32 发表评论

也就是搜索网站的职位，在cli中对应cli-boss search。

执行后，会生成一个cli_anything文件夹，里面有各种说明文件和源代码，包括了用户输入命令行后，背后需做的动作

paulwong 2026-04-12 15:11 发表评论

让ai保存成一个skill

直接保存成一个不依赖ai的可执行脚本

下次直接执行

还可以借助openclaw，进行定时自动化测试

整理成skill

如何使用这个skill

paulwong 2026-04-12 10:24 发表评论

paulwong 2026-04-06 07:37 发表评论

paulwong 2026-03-23 11:45 发表评论

https://skillhub.tencent.com/#categories

openclaw技能

https://clawhub.ai

paulwong 2026-03-21 03:42 发表评论

Field, Type, Index Name, Index Type, Index Parameters
id,auto id, Int64
vector, FloatVector(4096), vector, metric_type:COSINE
question, VarChar(5000)
answer, VarChar(5000)
#这三个字段vector, question, answer是必需的，且名字不能改

embedding:
  apiKey: "sk-xxxxxxx"
  model: "nvidia/llama-embed-nemotron-8b"
  path: "/v1/embeddings"
  serviceName: "llm-vllm-nvidia--llama-embed-nemotron-8b.internal.static"
  servicePort: 80
  type: "openai"
vector:
  apiKey: "empty-key"
  collectionID: "ai_higress_cache"
  serviceName: "my-milvus.static"
  servicePort: 80
  type: "milvus"
cacheKeyFrom: "messages.@reverse.0.content"
cacheKeyPrefix: "openai_gpt_oss_20b_"
cacheStreamValueFrom: "choices.0.delta.content"
cacheValueFrom: "choices.0.message.content"
returnResponseTemplate: |
  {"id":"from-cache","choices":[{"index":0,"message":{"role":"assistant","content":"%s"},"finish_reason":"stop"}],"model":"gpt-4o","object":"chat.completion","usage":{"prompt_tokens":0,"completion_tokens":0,"total_tokens":0}}
returnStreamResponseTemplate: |-
  data:{"id":"from-cache","choices":[{"index":0,"delta":{"role":"assistant","content":"%s"},"finish_reason":"stop"}],"model":"gpt-4o","object":"chat.completion","usage":{"prompt_tokens":0,"completion_tokens":0,"total_tokens":0}}
  data:[DONE]

paulwong 2026-03-11 18:07 发表评论