AI工作流

1. Promote blocked issues whose dependencies are now satisfied

agq promote

Build and Run

Container Build

With Podman (for Docker, replace with docker):

git clone https://github.com/lucasdicioccio/agents-exe
cd agents-exe/bundling
podman build -f Containerfile.build -t agents-exe
podman run -it --entrypoint=bash agents-exe:latest

Then inside the container:

agents-exe init
agents-exe tui

Quick Start

Getting started:

agents-exe init

will guide you into writing an agent.json and its tools directory.

Example: Using the OS Model

import System.Agents.OS
import System.Agents.OS.Core
import System.Agents.OS.Persistence

main :: IO ()
main = do
    -- Initialize the OS with SQLite persistence
    os <- initializeOS defaultConfig
    backend <- createPersistenceBackend (SqliteBackendType "./agents.db")
    
    -- Create a shared SQLite toolbox
    tbConfig <- createToolboxConfig SqliteToolboxType "shared-db" 
        (object ["path" .= "./data.db"])
    tbId <- createToolbox os tbConfig
    
    -- Create agents that share the toolbox
    agent1 <- createAgent os (testAgentConfig 
        { agentToolboxBindings = [tbId] })
    agent2 <- createAgent os (testAgentConfig 
        { agentToolboxBindings = [tbId] })
    
    -- Run conversations
    runOSM os $ do
        conv1 <- startConversation agent1 defaultConvConfig
        result1 <- runConversationTurn conv1 "Hello, agent 1!"
        
        conv2 <- startConversation agent2 defaultConvConfig
        result2 <- runConversationTurn conv2 "Hello, agent 2!"
        
        -- Both agents can access the same database!
        pure (result1, result2)

See docs/OS-API.md for the complete API reference.

Coding Style Guidelines

• Prefer qualified imports over prefixed/suffixed functions
• Keep prefixed/suffixed function names for helper variations
• Prefer passing multiple arguments
• Bundle arguments into Runtime or Props objects for long-running values or context
• No monad-tower, run in IO (library users can add their own)
• Use contravariant tracer to surface logs, be generous with traces

Global Options

These options apply to all commands:

Environment Variables

• EDITOR or GIT_EDITOR - Used for editing files (notably in the init command)

Advanced Configuration

Create an agents-exe.cfg.json in your project root (searched upward from current directory):

{
  "agentsConfigDir": "/some/dir",
  "agentsDirectories": ["./agents"],
  "agentsFiles": [],
  "agentsLogs": {
    "logJsonHttpEndpoint": "https://example.org/log-sink",
    "logRawPath": "agents-exe.raw",
    "logJsonPath": "logs.json",
    "logSessionsJsonPrefix": "sessions"
  }
}

Absent this file, agents load from ~/.config/agents-exe/default.

Command Reference

API Keys

Agents need an API key to authenticate against the LLM endpoint. By default agents-exe locates keys in ~/.config/agents-exe/secret-keys but you can override with --api-keys.

The keys file format:

{
  "keys": [
    {
      "id": "my-key-id",
      "value": "XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX"
    }
  ]
}

The id must match the apiKeyId field in the agent's JSON description.

CLI & Command Line

• optparse-applicative - Command-line option parsing

Lua Integration

• hslua - Lua bindings for Haskell
• hslua-marshalling - Lua marshalling utilities