03-planner-mode

Planner Mode

Planner mode gives an agent systematic planning capabilities. When the agent receives a complex or multi-step task, it will create a structured plan, track each step's progress, and verify completion before finishing.

This mode is ideal for tasks that require methodical execution: code generation with multiple files, research with several phases, or any workflow where step-by-step transparency matters.

See the Planner Mode Example for a complete, runnable implementation.

Enabling Planner Mode

import { Agent, TAgentMode } from '@smythos/sdk';

const agent = new Agent({
    name: 'Code Assistant',
    behavior: 'You are a senior developer who writes clean, well-tested code.',
    model: 'gpt-4o',
    mode: TAgentMode.PLANNER,
});

How It Works

When a user sends a complex request, the agent will:

1. Analyze the request and reason about the approach (in <thinking> tags)
2. Plan the steps needed (in <planning> tags)
3. Register the plan as tracked tasks via the internal _sre_Plan_Tasks skill
4. Execute each step sequentially, updating task status along the way
5. Verify that all tasks are completed before finishing

The agent communicates its reasoning transparently using <thinking> tags, and uses <planning> tags to outline its approach. These tags allow the LLM to separate internal reasoning from user-facing output.

Events

Planner mode emits events on the agent object so you can build UIs that reflect the agent's progress in real time.

TasksAdded

Fired when the agent creates a new plan.

agent.on('TasksAdded', (tasksList, allTasks) => {
    // tasksList: the newly added tasks
    // allTasks: all tasks in the planner (cumulative)
    for (const [id, task] of Object.entries(allTasks)) {
        console.log(`${task.status}: ${task.summary || task.description}`);
    }
});