Crate typesec_agent 

typesec-agent

Agent executor: typestate + capability-based access control in action.

This crate provides the high-level SecureAgent API that ties together:

• The typestate machine from typesec-core (unauthenticated → authenticated).
• Runtime policy checking via any [PolicyEngine].
• Typed capability acquisition: the only way to get a Capability<P, R> is through a successful policy check.
• Async task execution: the execute method requires a capability as proof.

Usage Pattern

// 1. Create agent with an engine — starts Unauthenticated.
let agent = SecureAgent::new(Arc::new(rbac_engine));

// 2. Authenticate — type transitions to Authenticated.
let agent = agent.authenticate(Credentials::new("agent:bot", "token"))?;

// 3. Request a capability — policy checked at runtime, cap minted on success.
let report = Report::new("reports/q1");
let cap: Capability<CanRead, Report> = agent.request_capability(&report).await?;

// 4. Execute — cap is required proof. No cap? Won't compile.
agent.execute(&cap, &report, |r| Box::pin(async move {
    println!("reading: {}", r.id);
    Ok(())
})).await?;

Re-exports

pub use agent::AgentBuilder;

pub use agent::SecureAgent;

pub use executor::TaskResult;

pub use tool::ProtectedTool;

pub use tool::ToolFuture;

pub use tool::ToolSpec;

Modules

agent

SecureAgent — the main agent struct wiring typestate + capabilities together.

executor

Task execution infrastructure.

tool

Capability-bound tool wrappers for agent and MCP-style tool execution.

Structs

CanDelete

Permission to delete a resource.

CanExecute

Permission to execute code or invoke actions on a resource.

CanRead

Permission to read a resource (non-sensitive).

CanWrite

Permission to write (create or update) a resource.

Capability

An unforgeable proof that subject subject holds permission P on resource R.

Credentials

Credentials used to authenticate an agent.

Traits

Permission

A marker trait for permissions.

Resource

A resource that can be protected by a Capability.