clawbox-types

Shared type definitions for the clawbox sandboxed agent execution service.

Overview

clawbox-types is the foundational crate in the clawbox ecosystem. It defines the core API types, tool manifests, sandbox policies, agent configuration, and host call interfaces used by every other clawbox crate. By centralizing these definitions, all crates stay in sync without circular dependencies.

Usage

use clawbox_types::{
    ExecuteRequest, ExecuteResponse, ExecutionMetadata,
    SandboxPolicy, Capabilities, NetworkCapabilities, ResourceLimits,
    AgentConfig, ToolManifest, ToolMeta,
};

// Build an execution request
let req = ExecuteRequest::new("web_search", serde_json::json!({"query": "rust wasm"}))
    .with_capabilities(Capabilities::new(
        NetworkCapabilities::new(vec!["api.example.com".into()])
    ));

// Configure an agent with builder pattern
let agent = AgentConfig::new("my-agent", "My Agent")
    .with_policy(SandboxPolicy::Container)
    .with_capabilities(
        Capabilities::new(NetworkCapabilities::new(vec!["api.github.com".into()]))
            .with_credential("github")
            .with_resources(ResourceLimits::new(60_000, 512))
    );

Features

• API types — ExecuteRequest, ExecuteResponse, ContainerSpawnRequest, HealthResponse, and more
• Tool manifests — ToolManifest with network, credential, and resource configurations
• Sandbox policies — WasmOnly, Container, and ContainerDirect isolation levels
• Agent types — AgentConfig, AgentInfo, AgentStatus for lifecycle management
• Resource limits — Timeout, memory, CPU, and output size budgets
• Host call trait — HostCallHandler for WASM-host RPC dispatch
• Credential patterns — Shared regex patterns for leak detection
• Builder patterns — Ergonomic construction for all major types
• #[non_exhaustive] — All structs and enums are non-exhaustive for semver safety

Architecture

License

MIT