Module prelude 

Common type re-exports.

Import everything with use echo_agent::prelude::*.

Re-exports

pub use crate::agent::AgentConfig;

pub use crate::agent::AgentRole;

pub use crate::agent::ReactAgent;

pub use crate::agent::ReactAgentBuilder;

pub use crate::agent::Runner;

pub use crate::agent::StructuredAgent;

pub use crate::config::AppConfig;

pub use crate::tools::builtin::think::ThinkTool;

pub use crate::workflow::WorkflowDefinition;

Structs

ActivateSkillTool

Tool for model-driven skill activation.

AnthropicClient

AuditCallback

基于 AgentCallback 的审计日志自动记录器

AuditEvent

Audit event

AuditFilter

审计查询过滤器

CancellationToken

A token which can be used to signal a cancellation request to one or more tasks.

ChatChunk

Streaming response chunk

ChatRequest

Chat request parameters

ChatResponse

Chat response

CircuitBreaker

Circuit Breaker

CircuitBreakerConfig

Circuit breaker configuration

CompressionInput

Compression pipeline input

CompressionOutput

Compression pipeline output

ConcurrentWorkflow

Concurrent workflow: all registered agents execute in parallel; results are merged via a merge function.

ContextManager

Context manager: maintains full conversation history and automatically triggers compression when tokens exceed the limit.

DagWorkflow

DAG workflow: nodes execute in topological order; independent nodes run concurrently.

DefaultPermissionPolicy

Default permission policy (retains backward compatibility)

DockerSandbox

Docker 容器沙箱

EmbeddingStore

向量增强 Store：在任意 Store 外包装余弦相似度语义检索

FileAuditLogger

文件审计日志记录器

FileCheckpointer

JSON file-based persistent Checkpointer

FileStore

JSON file-based persistent Store

FileSystemSkill

Filesystem skill

ForceCompressStats

Compression statistics returned by force_compress()

Graph

Compiled graph workflow (immutable)

GraphBuilder

Graph workflow builder

GraphResult

Graph execution result

GuardManager

Guard manager

HeuristicTokenizer

Heuristic Tokenizer that estimates token count using character weights.

HookRegistry

Registry of hooks from all activated skills.

HookResult

Result of executing a hook.

HookRule

A hook rule: a matcher pattern + one or more actions.

HooksDefinition

Complete hooks definition from a skill’s frontmatter.

HttpEmbedder

OpenAI-compatible HTTP embedding client

HybridCompressor

混合压缩：将多个 ContextCompressor 串联为有序管道

ImageFetchToolmedia

Image fetch tool

ImageUrl

Image URL or Base64 data

InMemoryAuditLogger

内存审计日志记录器

InMemoryCheckpointer

In-process memory Checkpointer, state is lost on restart, suitable for tests

InMemoryStore

In-process memory Store, no persistence, suitable for testing and short-lived use

JsonSchemaSpec

JSON Schema spec for structured output

K8sSandbox

Kubernetes Pod 沙箱

LlmConfig

LLM 运行时配置（依赖注入模式）

LlmGuardcontent-guard

LLM guard

LocalSandbox

本地沙箱执行器

Message

Conversation message

OllamaClient

OpenAiClient

OpenAI-compatible client.

PrepareResult

Result of ContextManager::prepare() — includes the prepared messages and optional compression stats if auto-compression was triggered.

PromptContext

Context for variable substitution and command execution.

ProviderFactory

Provider 工厂：通过配置字符串自动实例化 LLM 客户端

ReadSkillResourceTool

Tool for reading resource files from activated skill directories.

ResourceLimits

Resource limits

RetryPolicy

Unified retry policy configuration

RuleGuardcontent-guard

Rule guard

RuleGuardBuildercontent-guard

RuleGuard builder

RunSkillScriptTool

Tool for executing scripts from activated skill directories.

SandboxCommand

Sandbox execution command

SandboxManager

沙箱管理器：统一调度三层沙箱

SandboxPolicy

沙箱策略：根据命令和上下文决定所需隔离级别

SandboxResult

Execution result

SequentialWorkflow

Sequential workflow: each agent runs in registration order; the former’s output flows into the latter’s input.

SharedState

Shared state for graph workflows

ShellSkill

Shell command execution skill

SimpleTokenizer

Simple Tokenizer: byte_count / 4 + 1 (backward compatible with old behavior)

SkillContent

Full skill content returned when a skill is activated (Tier 2).

SkillDescriptor

Lightweight skill metadata loaded at discovery time (Tier 1).

SkillInfo

Metadata snapshot for an installed code-based skill.

SkillLoader

Multi-scope skill loader with agentskills.io-compliant parsing.

SkillRegistry

Central skill lifecycle manager.

SkillResourceEntry

A bundled resource file discovered in the skill directory.

SlidingWindowCompressor

滑动窗口压缩：保留最近 window_size 条非 system 消息，裁掉更早的部分。

SnapshotManager

管理状态快照的环形缓冲区

SqliteStoresqlite

SQLite persistent Store with FTS5 full-text search and optional vector search

StateSnapshot

单次状态快照，记录某一时刻的完整对话历史

StepOutput

Detailed output of a single step execution

StoreItem

A single record in the Store

SummaryCompressor

LLM 摘要压缩策略。

ToolCall

A single tool call initiated by the LLM

ToolDefinition

Tool definition sent to the LLM

ToolExecutionConfig

Tool execution config: timeout, retry, concurrency

ToolResult

工具执行结果

WebFetchToolweb

Web page fetching tool

WebFetchToolEnhancedmedia

Enhanced web fetch tool with image download support

WebSearchToolweb

Web search tool

WorkflowOutput

Complete output of a workflow execution

Enums

AgentEvent

Events produced during Agent execution

AuditEventType

Audit event types

ContentPart

A single component of message content (multimodal)

DiscoveryScope

Where to scan for skills.

GuardDirection

Guard check direction

GuardResult

Guard check result

HookAction

A single hook action.

HookEvent

When a hook fires relative to tool execution.

IsolationLevel

Isolation level

LlmProvider

LLM 供应商类型

MessageContent

Message content: compatible with both plain text and multimodal parts forms.

PermissionDecision

Permission decision

ResponseFormat

Response format control

SecurityLevel

安全级别

SkillResourceKind

Classification of a bundled resource file.

SkillSource

Source of a skill (affects security policy).

SnapshotPolicy

自动快照策略

StepType

Parsed step type from LLM response

ToolPermission

Tool permission types

WorkflowEvent

Step-by-step events emitted during workflow execution.

Traits

Agent

Unified Agent execution interface

AgentCallback

Agent lifecycle callback interface

AuditLogger

审计日志记录器 trait

Checkpointer

Persistence interface for short-term conversation memory

ContextCompressor

Unified interface for all compression strategies (async, supports dyn trait object)

Embedder

Text embedding interface: maps text to dense float vectors

Guard

Guard trait

LlmClient

LLM client unified interface

PermissionPolicy

Permission policy trait

SandboxExecutor

Sandbox executor unified interface.

Skill

Agent skill – a higher-level capability unit that bundles related tools with an optional system-prompt injection.

Store

Unified storage interface for long-term memory

Tokenizer

Token counter abstraction

Tool

Tool interface trait

Workflow

Unified workflow execution interface

Functions

default_summary_prompt

使用内置中文模板生成默认摘要提示词。

shared_agent

Wrap arbitrary impl Agent as a SharedAgent

with_retry

Execute an async operation with the given retry policy (all errors are considered retryable).

with_retry_if

Execute an async operation with the given retry policy.

Type Aliases

AgentBuilder

Alias for backward compatibility.

Result

Convenience Result alias

SharedAgent

Shareable agent handle for safe access across async tasks

ToolParameters

Tool parameter type