Module testing 

Testing utilities for the managed agent runtime.

This module provides deterministic test doubles for the managed agent runtime pipeline. These are not mocks — they implement the real traits and exercise the full runtime pipeline (parking, checkpoints, replay, event mapping). Only the LLM provider API call is replaced with pre-scripted deterministic responses.

Architecture

ScriptedLlm (deterministic responses)
  │
  ▼
Full runtime pipeline (SessionLoop, CheckpointManager, ToolParkingLot, etc.)
  │
  ▼
SessionEvent stream (byte-identical assertions possible)

Usage

use adk_managed::testing::{ScriptedLlm, ScriptedTurn, ScriptedToolCall};
use serde_json::json;

let turns = vec![
    ScriptedTurn {
        text: Some("Hello! How can I help you?".to_string()),
        tool_calls: vec![],
    },
    ScriptedTurn {
        text: None,
        tool_calls: vec![ScriptedToolCall {
            name: "web_search".to_string(),
            input: json!({"query": "rust async"}),
            id: Some("tc_001".to_string()),
        }],
    },
];

let llm = ScriptedLlm::new("scripted-model", turns);
// Use llm in place of any Arc<dyn Llm> in the runtime pipeline

Structs

ScriptedLlm

A deterministic LLM double with pre-scripted responses.

ScriptedToolCall

A scripted tool call within a ScriptedTurn.

ScriptedTurn

A pre-scripted turn that the ScriptedLlm will return.