使用示例

LibreFang 常用场景的完整示例。

场景 1: 客户服务 Agent

需求

7x24 小时自动回复客户咨询
接入 Telegram 和 Discord
记住客户历史对话
需要人工审批复杂问题

Agent 清单

# customer-support.toml
name = "customer-support"
version = "1.0.0"
description = "Customer support agent with approval workflow"
module = "builtin:chat"

[model]
provider = "groq"
model = "llama-3.3-70b-versatile"

[system_prompt]
prompt = """你是一个专业的客服 agent。你的职责是：
1. 礼貌回复客户咨询
2. 提供准确的产品信息
3. 复杂问题转人工
4. 保持专业和友好"""

[capabilities]
tools = ["file_read", "web_fetch", "memory_recall", "memory_store"]
memory_read = ["customer:*"]
memory_write = ["customer:*"]
network = ["*"]

[resources]
max_llm_tokens_per_hour = 50000

[approval]
required = true
channels = ["telegram"]
keywords = ["refund", "complaint", "manager"]

通道配置

[telegram]
bot_token_env = "TELEGRAM_BOT_TOKEN"
allowed_users = ["*"]

[discord]
bot_token_env = "DISCORD_BOT_TOKEN"
guild_ids = ["123456789"]

[channels.customer-support]
system_prompt = "你是一个客服 agent，请礼貌专业地回复客户。"
rate_limit = 20

场景 2: 每日新闻摘要

需求

每天早上 8 点自动运行
抓取多个新闻源
生成摘要
发送到 Slack

工作流

{
  "name": "daily-news-digest",
  "version": "1.0.0",
  "trigger": {
    "type": "schedule",
    "cron": "0 8 * * *"
  },
  "steps": [
    {
      "id": "fetch-tech-news",
      "type": "agent",
      "agent": "researcher",
      "input": "Fetch latest tech news from Hacker News and TechCrunch",
      "output_var": "tech_news"
    },
    {
      "id": "fetch-ai-news",
      "type": "agent",
      "agent": "researcher",
      "input": "Fetch latest AI news from AI News and OpenAI blog",
      "output_var": "ai_news"
    },
    {
      "id": "summarize",
      "type": "agent",
      "agent": "writer",
      "input": "Create a brief summary of these news items: {{tech_news}} and {{ai_news}}",
      "output_var": "summary"
    },
    {
      "id": "send-slack",
      "type": "notification",
      "channel": "slack",
      "message": "📰 Daily News Digest\n\n{{summary}}"
    }
  ]
}

场景 3: 代码审查助手

需求

监听 GitHub PR
自动进行代码审查
提供改进建议

Agent 清单

# code-reviewer.toml
name = "code-reviewer"
version = "1.0.0"
description = "Automated code review agent"
module = "builtin:chat"

[model]
provider = "anthropic"
model = "claude-sonnet-4-20250514"

[system_prompt]
prompt = """你是一个专业的代码审查专家。你的职责是：
1. 检查代码质量和风格
2. 识别潜在 bug 和安全问题
3. 提供具体的改进建议
4. 保持建设性和专业"""

[capabilities]
tools = ["file_read", "web_fetch", "git"]
memory_read = ["review:*"]
memory_write = ["review:*"]

[schedule]
enabled = true
cron = "0 * * * *"

Webhook 触发器

{
  "type": "webhook",
  "path": "/webhook/github",
  "events": ["pull_request"]
}

场景 4: 销售潜在客户挖掘

需求

每天自动挖掘潜在客户
丰富公司信息
评分和排序
发送到 CRM

Hand 配置

[lead]
icp = "B2B SaaS companies, 50-500 employees, Series A+"
search_sources = ["linkedin", "crunchbase", "hunter"]
score_criteria = [
  "funding_stage: A+ = 30",
  "employee_count: 50-500 = 20",
  "tech_stack: contains(ai|ml) = 25",
  "job_postings: recent = 15"
]
deduplicate = true
crm_integration = "hubspot"

[[lead.quality_thresholds]]
score = 80
action = "auto_add_to_crm"

[[lead.quality_thresholds]]
score = 50
action = "require_approval"

工作流

{
  "name": "lead-generation",
  "trigger": {
    "type": "schedule",
    "cron": "0 9 * * *"
  },
  "steps": [
    {
      "id": "discover",
      "type": "agent",
      "agent": "lead",
      "input": "Find 50 companies matching our ICP"
    },
    {
      "id": "enrich",
      "type": "agent",
      "agent": "collector",
      "input": "Enrich company data: {{discover.companies}}"
    },
    {
      "id": "score",
      "type": "agent",
      "agent": "classifier",
      "input": "Score leads: {{enrich.leads}}"
    },
    {
      "id": "notify",
      "type": "notification",
      "channel": "telegram",
      "message": "Generated {{score.high_quality_count}} qualified leads"
    }
  ]
}

场景 5: 内容创作助手

需求

定期生成社交媒体内容
支持多个平台
审批工作流

Twitter Hand 配置

[twitter]
content_formats = ["thread", "hot_take", "question"]
topics = ["ai", "technology", "productivity"]
approval_queue = true
posting_times = ["9:00", "12:00", "18:00"]
max_posts_per_day = 5

[twitter.linkedin]
enabled = true
posting_times = ["10:00"]

内容审批工作流

{
  "name": "content-approval",
  "steps": [
    {
      "id": "generate",
      "type": "agent",
      "agent": "twitter",
      "input": "Generate 3 tweet ideas about AI"
    },
    {
      "id": "create",
      "type": "agent",
      "agent": "writer",
      "input": "Write a thread based on: {{generate.ideas[0]}}"
    },
    {
      "id": "approve",
      "type": "approval",
      "channels": ["telegram"],
      "timeout": "24h"
    },
    {
      "id": "publish",
      "type": "notification",
      "channel": "twitter",
      "message": "{{create.content}}"
    }
  ]
}

场景 6: 数据库助手

需求

回答数据库相关问题
执行只读查询
生成 SQL

Agent 清单

# db-assistant.toml
name = "db-assistant"
version = "1.0.0"
module = "builtin:chat"

[model]
provider = "anthropic"
model = "claude-sonnet-4-20250514"

[system_prompt]
prompt = """你是一个数据库助手。你可以：
1. 回答关于数据库设计的问题
2. 编写和优化 SQL 查询
3. 解释查询执行计划
注意：不要执行任何修改数据的操作。"""

[capabilities]
tools = ["file_read", "postgres_query"]
memory_read = ["schema:*"]

[security]
read_only = true
allowed_tables = ["users", "orders", "products", "analytics_*"]
max_rows = 1000

场景 7: 个人知识助手

需求

管理和检索个人笔记
语义搜索
自动整理

Agent 清单

# knowledge-assistant.toml
name = "knowledge-assistant"
version = "1.0.0"
module = "builtin:chat"

[model]
provider = "groq"
model = "llama-3.3-70b-versatile"

[capabilities]
tools = ["file_read", "file_write", "memory_recall", "memory_store"]
memory_read = ["knowledge:*", "notes:*", "bookmarks:*"]
memory_write = ["knowledge:*", "notes:*"]

[memory]
decay_rate = 0.02

使用示例

# 存储笔记
librefang memory set --key "notes:meeting-2025-01-15" --value "Discussed project roadmap"

# 语义搜索
librefang memory search "project planning"

# 知识图谱查询
librefang kg query --from "person:alice" --relation "works_with"

场景 8: CI/CD 监控助手

需求

监控 GitHub Actions
失败时告警
提供调试建议

配置

[github]
webhook_secret_env = "GITHUB_WEBHOOK_SECRET"

[[triggers]]
type = "webhook"
path = "/webhook/github"
events = ["workflow_run.completed", "workflow_run.failed"]

[notifications]
channels = ["slack", "telegram"]

工作流

{
  "name": "cicd-monitor",
  "trigger": {
    "type": "event",
    "event": "workflow_run.failed"
  },
  "steps": [
    {
      "id": "analyze",
      "type": "agent",
      "agent": "debugger",
      "input": "Analyze failed workflow: {{event.workflow_url}}"
    },
    {
      "id": "notify",
      "type": "notification",
      "channel": "slack",
      "message": "⚠️ Workflow Failed: {{event.workflow_name}}\n\n{{analyze.suggestion}}"
    }
  ]
}

场景 9: 智能家居控制

需求

通过聊天控制智能设备
定时任务
场景自动化

Agent 清单

# smart-home.toml
name = "smart-home"
version = "1.0.0"
module = "builtin:chat"

[model]
provider = "groq"
model = "llama-3.1-8b-instant"

[capabilities]
tools = ["shell_execute", "http_request"]

[allowed_devices]
lights = ["living_room", "bedroom", "kitchen"]
thermostat = ["main"]
locks = ["front_door", "back_door"]
cameras = ["front_porch", "backyard"]

命令示例

用户: "帮我关掉客厅的灯"
Agent: 已关闭客厅的灯 ✓

用户: "设置下午 6 点的温度为 22 度"
Agent: 已设置定时任务：18:00 温度调节至 22°C ✓

用户: "如果有人在前门摄像头出现，就给我发 Telegram"
Agent: 已创建自动化规则：摄像头检测 → Telegram 通知 ✓

场景 10: 会议助手

需求

安排会议
发送邀请
会议纪要

Agent 清单

# meeting-assistant.toml
name = "meeting-assistant"
version = "1.0.0"
module = "builtin:chat"

[model]
provider = "groq"
model = "llama-3.3-70b-versatile"

[capabilities]
tools = ["calendar", "email", "file_write", "memory_recall"]
memory_read = ["meeting:*", "contacts:*"]
memory_write = ["meeting:*"]

[calendar]
provider = "google"
timezone = "Asia/Shanghai"

[email]
smtp_host = "smtp.gmail.com"

功能

自然语言安排会议
自动查找空闲时间
发送日历邀请
会后生成纪要
跟踪行动项目