Crate enact_runner 

enact-runner — The Robust Agent Loop for Enact

This crate fills the gap between enact-core’s deterministic kernel and the applications (CLI, API) that assume the Rust layer handles retries, context compaction, and multi-format tool call parsing.

Architecture

┌─────────────────────────────────────────────────────┐
│  CLI / API / Web (thin clients)                     │
└──────────────────────┬──────────────────────────────┘
                       │ Execution Intent
                       ▼
┌─────────────────────────────────────────────────────┐
│  enact-runner (THIS CRATE)                          │
│  ┌─────────────────────────────────────────────┐    │
│  │  AgentRunner::run()                         │    │
│  │  • Multi-iteration tool call loop           │    │
│  │  • Retry with exponential backoff           │    │
│  │  • Auto context compaction                  │    │
│  │  • Multi-format parsing (JSON/XML/MD)       │    │
│  │  • Checkpoint at intervals                  │    │
│  │  • Stream event emission                    │    │
│  └─────────────────────────────────────────────┘    │
└──────────────────────┬──────────────────────────────┘
                       │ Callable::run()
                       ▼
┌─────────────────────────────────────────────────────┐
│  enact-core kernel (deterministic DAG engine)       │
└─────────────────────────────────────────────────────┘

Quick Start

use enact_runner::{AgentRunner, RunnerConfig, DefaultAgentRunner};

// Create runner with default config
let mut runner = DefaultAgentRunner::default_new();

// Or configure for long-running tasks
let mut runner = DefaultAgentRunner::with_config(RunnerConfig::long_running());

// Run an agent through the robust loop
// let outcome = runner.run(&my_callable, "user input").await?;

Re-exports

pub use approval::checker_from_approval_config;

pub use approval::checker_from_config;

pub use approval::AlwaysApprove;

pub use approval::AlwaysDeny;

pub use approval::ApprovalChecker;

pub use approval::ApprovalPrompter;

pub use approval::AskApprovalChecker;

pub use approval::AskOnceApprovalChecker;

pub use approval::CliPrompter;

pub use approval::PatternApprovalChecker;

pub use approval::PolicyWithOverrides;

pub use commands::dispatch as dispatch_command;

pub use commands::load_commands;

pub use commands::load_commands_for_run;

pub use commands::CommandDef;

pub use compaction::HistoryMessage;

pub use config::ObservabilityConfig;

pub use config::RetryConfig;

pub use config::RunnerConfig;

pub use hooks::load_global_hooks;

pub use hooks::HookRegistry;

pub use loop_driver::AgentRunner;

pub use loop_driver::DefaultAgentRunner;

pub use loop_driver::LoopOutcome;

pub use parser::ParseResult;

pub use parser::ParsedToolCall;

pub use parser::ToolCallFormat;

pub use registry::LiveSessionInfo;

pub use registry::LiveSessionRegistry;

pub use retry::classify_error;

pub use retry::ErrorKind;

pub use retry::RetryHandler;

pub use session::SessionDetails;

pub use session::SessionEvent;

pub use session::SessionQuery;

pub use session::SessionReader;

pub use session::SessionRecorder;

pub use session::SessionStatus;

pub use session::SessionSummary;

pub use session::SortOrder;

Modules

approval

Human-in-the-loop approval — check before tool execution.

commands

Slash commands — user-invokable commands from ~/.enact/commands/, .enact/commands/, and plugins.

compaction

Context compaction for long-running agent loops

config

Runner configuration

hooks

Hook configuration and registry — load hooks from ~/.enact/hooks.yaml, plugins, and agent YAML.

loop_driver

Core agent loop driver

parser

Multi-format tool call parser

registry

Live Session Registry — tracks currently running sessions for message routing

retry

Retry handler with exponential backoff and error classification

session

Session recorder and reader — one JSONL file per run under agents//sessions/ and optionally projects//sessions/.