Crate a3s_code_core 

A3S Code Core Library

Embeddable AI agent library with tool execution capabilities. This crate contains all business logic extracted from the A3S Code agent, enabling direct Rust API usage as an embedded library.

Quick Start

use a3s_code_core::{Agent, AgentEvent};

// From a config file path (.hcl or .json)
let agent = Agent::new("agent.hcl").await?;

// Create a workspace-bound session
let session = agent.session("/my-project", None)?;

// Non-streaming
let result = session.send("What files handle auth?", None).await?;
println!("{}", result.text);

// Streaming (AgentEvent is #[non_exhaustive])
let (mut rx, _handle) = session.stream("Refactor auth", None).await?;
while let Some(event) = rx.recv().await {
    match event {
        AgentEvent::TextDelta { text } => print!("{text}"),
        AgentEvent::End { .. } => break,
        _ => {} // required: #[non_exhaustive]
    }
}

Architecture

Agent (facade — config-driven, workspace-independent)
  +-- LlmClient (Anthropic / OpenAI)
  +-- CodeConfig (HCL / JSON)
  +-- SessionManager (multi-session support)
        |
        +-- AgentSession (workspace-bound)
              +-- AgentLoop (core execution engine)
              |     +-- ToolExecutor (14 tools: 11 builtin + 3 skill discovery)
              |     +-- LlmPlanner (JSON-structured planning)
              |     +-- HITL Confirmation
              +-- HookEngine (8 lifecycle events)
              +-- Security (sanitizer, taint, injection detection, audit)
              +-- Memory (episodic, semantic, procedural, working)
              +-- MCP (JSON-RPC 2.0, stdio + HTTP+SSE)
              +-- Cost Tracking / Telemetry

Re-exports

pub use agent::AgentConfig;

pub use agent::AgentEvent;

pub use agent::AgentLoop;

pub use agent::AgentResult;

pub use agent_api::Agent;

pub use agent_api::AgentSession;

pub use agent_api::BtwResult;

pub use agent_api::SessionOptions;

pub use agent_api::ToolCallResult;

pub use agent_teams::AgentExecutor;

pub use agent_teams::AgentTeam;

pub use agent_teams::TeamConfig;

pub use agent_teams::TeamMember;

pub use agent_teams::TeamMemberOptions;

pub use agent_teams::TeamMessage;

pub use agent_teams::TeamRole;

pub use agent_teams::TeamRunResult;

pub use agent_teams::TeamRunner;

pub use agent_teams::TeamTaskBoard;

pub use commands::CommandAction;

pub use commands::CommandContext;

pub use commands::CommandOutput;

pub use commands::CommandRegistry;

pub use commands::CronCancelCommand;

pub use commands::CronListCommand;

pub use commands::LoopCommand;

pub use commands::SlashCommand;

pub use config::CodeConfig;

pub use config::ModelConfig;

pub use config::ModelCost;

pub use config::ModelLimit;

pub use config::ModelModalities;

pub use config::ProviderConfig;

pub use error::CodeError;

pub use error::Result;

pub use hooks::HookEngine;

pub use llm::AnthropicClient;

pub use llm::Attachment;

pub use llm::ContentBlock;

pub use llm::ImageSource;

pub use llm::LlmClient;

pub use llm::LlmResponse;

pub use llm::Message;

pub use llm::OpenAiClient;

pub use llm::TokenUsage;

pub use orchestrator::AgentSlot;

pub use plugin::Plugin;

pub use plugin::PluginContext;

pub use plugin::PluginManager;

pub use plugin::SkillPlugin;

pub use queue::ExternalTask;

pub use queue::ExternalTaskResult;

pub use queue::LaneHandlerConfig;

pub use queue::SessionCommand;

pub use queue::SessionLane;

pub use queue::SessionQueueConfig;

pub use queue::SessionQueueStats;

pub use queue::TaskHandlerMode;

pub use sandbox::SandboxConfig;

pub use scheduler::CronScheduler;

pub use scheduler::ScheduledFire;

pub use scheduler::ScheduledTask;

pub use scheduler::ScheduledTaskInfo;

pub use session::SessionConfig;

pub use session::SessionManager;

pub use session::SessionState;

pub use session_lane_queue::SessionLaneQueue;

pub use skills::builtin_skills;

pub use skills::Skill;

pub use skills::SkillKind;

pub use task::manager::TaskEvent;

pub use task::AgentProgress;

pub use task::Coordinator;

pub use task::IdlePhase;

pub use task::IdleTask;

pub use task::IdleTurn;

pub use task::ProgressTracker;

pub use task::Task;

pub use task::TaskId;

pub use task::TaskManager;

pub use task::TaskResult;

pub use task::TaskStatus;

pub use task::TaskTokenUsage;

pub use task::TaskType;

pub use task::ToolActivity;

pub use tool_search::ToolIndex;

pub use tool_search::ToolMatch;

pub use tool_search::ToolSearchConfig;

pub use tools::ToolContext;

pub use tools::ToolExecutor;

pub use tools::ToolResult;

Modules

agent

Agent Loop Implementation

agent_api

Agent Facade API

agent_teams

Agent Teams — Peer-to-peer multi-agent coordination

commands

Slash Commands — Interactive session commands

config

Configuration module for A3S Code

context

Context Provider Extension Point

error

Typed error enum for A3S Code Core

file_history

File version history tracking

git

Git tool — Uses system git with auto-installation support

hitl

Human-in-the-Loop (HITL) confirmation mechanism

hooks

Hooks System for A3S Code Agent

llm

LLM client abstraction layer

mcp

MCP (Model Context Protocol) Support

memory

Memory and learning system for the agent.

orchestrator

Agent Orchestrator - 主子智能体协调器

permissions

Permission system for tool execution control

planning

Planning, Goal Tracking, and Task Management

plugin

Plugin System for A3S Code

queue

Per-session command queue with lane-based priority scheduling

sandbox

Sandbox integration for bash tool execution.

scheduler

Session-scoped prompt scheduler — backs /loop, /cron-list, /cron-cancel.

security

Security Module

session

Session management

session_lane_queue

Session Lane Queue - a3s-lane backed command queue

skills

Skill System

store

Session persistence layer

task

Task Lifecycle Management

telemetry

Telemetry Module (Core)

tool_search

Tool Search — Semantic tool matching for dynamic MCP tool loading

tools

Extensible Tool System

undercover

Undercover mode for safe public repository operations

Structs

AgentDefinition

Agent definition

AgentRegistry

Agent registry for managing agent definitions

MetricsSnapshot

Snapshot of all metrics at a point in time

SystemPromptSlots

Slot-based system prompt customization with intent-based style selection.

Enums

AgentStyle

Agent style — determines which system prompt template is used.

DetectionConfidence

Detection confidence level for style detection.

PlanningMode

Planning mode — controls when planning phase is used.

Constants

AGENT_VERIFICATION

Verification agent — adversarial specialist that tries to break code

AGENT_VERIFICATION_RESTRICTIONS

Tool restrictions for verification agent

INTENT_CLASSIFY_SYSTEM

System prompt for LLM-based intent classification

PROMPT_SUGGESTION

Prompt suggestion service with filtering rules

SESSION_MEMORY_TEMPLATE

Session memory template with structured sections

SUBAGENT_EXPLORE

Explore subagent — read-only codebase exploration

SUBAGENT_PLAN

Plan subagent — read-only planning and analysis

SUBAGENT_SUMMARY

Summary subagent — summarize conversation key points

SUBAGENT_TITLE

Title subagent — generate concise conversation title

UNDERCOVER_INSTRUCTIONS

Undercover mode instructions for commit/PR prompts

Functions

load_agents_from_dir

Load all agent definitions from a directory