Crate clawbox_types 

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

Module	Contents
api	HTTP request/response types (ExecuteRequest, etc.)
policy	SandboxPolicy, Capabilities, ResourceLimits
manifest	ToolManifest, ToolMeta, network/credential config
agent	AgentConfig, AgentInfo, lifecycle types
host	HostCallHandler trait for WASM-host RPC
patterns	Shared credential detection regex patterns

License

MIT

Re-exports

pub use agent::AgentConfig;

pub use agent::AgentInfo;

pub use agent::AgentStatus;

pub use agent::LifecycleConfig;

pub use agent::WorkspaceConfig;

pub use api::ApiError;

pub use api::ComponentHealth;

pub use api::ContainerInfo;

pub use api::ContainerSpawnRequest;

pub use api::ContainerStatus;

pub use api::ExecuteRequest;

pub use api::ExecuteResponse;

pub use api::ExecutionMetadata;

pub use api::ExecutionStatus;

pub use api::HealthComponents;

pub use api::HealthResponse;

pub use api::ResourceUsage;

pub use api::SanitizationReport;

pub use host::HostCallHandler;

pub use manifest::ToolCredentialConfig;

pub use manifest::ToolManifest;

pub use manifest::ToolMeta;

pub use manifest::ToolNetworkConfig;

pub use manifest::ToolResourceConfig;

pub use policy::Capabilities;

pub use policy::NetworkCapabilities;

pub use policy::ResourceLimits;

pub use policy::SandboxPolicy;

Modules

agent

Agent-level types for container orchestration.

api

API request and response types for the clawbox HTTP interface.

host

Host call handler trait for WASM-host RPC dispatch.

manifest

Tool capability manifests — define what a WASM tool can do.

patterns

Shared credential and injection patterns used across clawbox crates.

policy

Sandbox policies defining the security envelope for executions.

Constants

DEFAULT_MAX_CONCURRENT

Default maximum concurrent connections, shared across policy and manifest.