Agent 模板目录

librefang spawn orchestrator
librefang spawn coder
librefang spawn --template agents/writer/agent.toml

curl -X POST http://localhost:4545/api/agents \
  -H "Content-Type: application/json" \
  -d '{"template": "coder"}'

curl -X POST http://localhost:4545/api/agents \
  -H "Content-Type: application/json" \
  -d '{"template": "writer", "model": "claude-sonnet-4-20250514"}'

curl -X POST http://localhost:4545/api/agents/{id}/message \
  -H "Content-Type: application/json" \
  -d '{"content": "Write unit tests for the auth module"}'

# 必填字段
name = "my-agent"
version = "0.1.0"
description = "What this agent does in one sentence."
author = "your-name"
module = "builtin:chat"

# 可选元数据
tags = ["tag1", "tag2"]

# 模型配置（必填）
[model]
provider = "default"                 # 继承 config.toml 中的默认供应商
model = "default"                    # 继承 config.toml 中的默认模型
max_tokens = 4096                    # 每次响应的最大输出 token 数
temperature = 0.3                    # 创造性（0.0 = 确定性，1.0 = 创造性）
system_prompt = """Your agent's personality, capabilities, and instructions go here.
Be specific about what the agent should and should not do."""

# To use a specific provider instead of default, set the provider and model explicitly:
# provider = "gemini"
# model = "gemini-2.5-flash"
# api_key_env = "GEMINI_API_KEY"

# Optional fallback model (used when primary is unavailable)
# [[fallback_models]]
# provider = "groq"
# model = "llama-3.3-70b-versatile"
# api_key_env = "GROQ_API_KEY"

# 可选调度（用于自主/后台 Agent）
[schedule]
periodic = { cron = "every 5m" }                                     # 定期执行
# continuous = { check_interval_secs = 120 }                         # 持续循环
# proactive = { conditions = ["event:agent_spawned"] }               # 事件触发

# 资源限制
[resources]
max_llm_tokens_per_hour = 150000    # 每小时 Token 预算

# 能力授权（最小权限原则）
[capabilities]
tools = ["file_read", "file_write", "file_list", "shell_exec",
         "memory_store", "memory_recall", "web_fetch",
         "agent_send", "agent_list", "agent_spawn", "agent_kill"]
network = ["*"]                     # 网络访问模式
memory_read = ["*"]                 # Agent 可读取的记忆命名空间
memory_write = ["self.*"]           # Agent 可写入的记忆命名空间
agent_spawn = true                  # 此 Agent 是否可以生成其他 Agent
agent_message = ["*"]               # 可以向哪些 Agent 发送消息
shell = ["python *", "cargo *"]     # 允许的 Shell 命令模式（白名单）

# 命名共享工作区（可选）。
# 多个 Agent 可以声明同一路径 — 它们共享该目录而不会产生 identity 文件冲突。
# 路径相对于 workspaces_dir。
# [workspaces]
# library = { path = "shared/library", mode = "rw" }
# archive = { path = "shared/archive", mode = "r"  }

workspaces/agents/librarian/
├── .identity/          ← identity 文件（由内核管理）
│   ├── SOUL.md         ← 人格与使命
│   ├── IDENTITY.md     ← 视觉身份（emoji、颜色、风格）
│   ├── USER.md         ← 学到的用户偏好
│   ├── TOOLS.md        ← 工具与环境说明
│   ├── MEMORY.md       ← 长期记忆
│   ├── AGENTS.md       ← 行为规范
│   └── BOOTSTRAP.md    ← 首次运行协议
├── sessions/           ← 对话历史
├── memory/             ← 每日记忆日志
├── data/               ← Agent 专属数据
├── output/             ← 生成的输出文件
├── skills/             ← Agent 专属技能
└── logs/               ← Agent 日志

# librarian/agent.toml
[workspaces]
library = { path = "shared/library", mode = "rw" }

# researcher/agent.toml
[workspaces]
library = { path = "shared/library", mode = "r" }   # 只读

## Shared Workspaces
- @library → /home/you/.librefang/workspaces/shared/library (read-write)

# agent.toml
[workspaces]
vault = { mount = "/Users/me/Documents/Obsidian", mode = "r" }

# config.toml —— 启用上面示例所需
allowed_mount_roots = [
  "/Users/me/Documents",
]

# agent.toml
cache_context = true   # 默认 false（每回合重读）

# agent.toml —— 关闭懒加载；每回合发送所有 tool schema
lazy_tools = false

# 按模板名称生成
librefang spawn coder

# 使用自定义名称生成
librefang spawn coder --name "backend-coder"

# 从 TOML 文件路径生成
librefang spawn --template agents/custom/my-agent.toml

# 列出运行中的 Agent
librefang agents

# 发送消息
librefang message <agent-id> "写一个解析 TOML 文件的函数"

# 终止 Agent
librefang kill <agent-id>

# 从模板生成
POST /api/agents
{"template": "coder"}

# 生成并覆盖配置
POST /api/agents
{"template": "coder", "name": "backend-coder", "model": "deepseek-chat"}

# 发送消息
POST /api/agents/{id}/message
{"content": "实现 auth 模块"}

# WebSocket（流式）
WS /api/agents/{id}/ws

# 列出 Agent
GET /api/agents

# 删除 Agent
DELETE /api/agents/{id}

# 通过 OpenAI 兼容端点使用任意 Agent
POST /v1/chat/completions
{
  "model": "librefang:coder",
  "messages": [{"role": "user", "content": "用 Rust 写一个 HTTP 服务器"}],
  "stream": true
}

# 列出可用模型
GET /v1/models

用户："构建一个带测试和文档的 REST API"

Orchestrator：
1. agent_send(coder, "实现 REST API 端点")
2. agent_send(test-engineer, "为这些端点编写集成测试")
3. agent_send(doc-writer, "为 API 端点撰写文档")
4. 将所有结果综合成最终报告

[model]
provider = "groq"                    # 或 "gemini"、"openai"、"anthropic"、"deepseek"
model = "llama-3.3-70b-versatile"    # 对应供应商的模型标识符
api_key_env = "GROQ_API_KEY"         # 保存 API 密钥的环境变量

tool_allowlist = ["web_search", "web_fetch", "file_read", "file_write"]
tool_blocklist = ["shell_exec"]
tools_disabled = false             # 显式总开关 —— 比所有别的设置优先
allowed_plugins = ["truncate", "redact"]   # 插件名;空 = 所有可用插件
mcp_servers    = ["github", "linear"]     # MCP server 名;空 = 所有已连接 server
skills_disabled = false            # skill 的显式总开关

generate_identity_files = true     # 首次 spawn 时创建 SOUL.md / IDENTITY.md / TOOLS.md
inherit_parent_context  = true     # 子 agent 默认继承父 agent context
auto_evolve             = true     # 每轮后台跑 skill-evolution review
is_hand                 = false    # 由 Hand spawn 时自动置位
pinned_model            = "anthropic/claude-sonnet-4"   # Stable kernel 模式下锁定模型

[channel_overrides]
dm_policy = "always"
group_policy = "trigger_only"
group_trigger_patterns = ["(?i)\\bmy-bot\\b"]

[exec_policy]
shell_allowlist = ["git", "cargo"]
deny = ["rm -rf"]

show_progress = true
response_format = "json_object"   # 或 { type = "json_schema", schema = "..." }
web_search_augmentation = "auto"  # "off" | "auto" | "always"

[[context_injection]]
name = "rules"
content = "保持简洁。永远不用 emoji。"
position = "system"