feat: allow user to modify a plan prepared for complex prompts by TodoOrchestrator

[ccenzura]

Feature description

Give user opportunity to have an impact on proposed plan before start realizing the tasks from the plan.

Motivation

Right now, there is no option to modify prepared plan based on the first prompt. Just after print of all tasks Agent starts working on the first one and so on.

Additional context

I would like to create tutorial with example if the Agent for preparing specification (Spec Driven Development) for a new projects based on vibe coding.

[ccenzura]

ToDoOrchestrator: Hierarchical Task Management with Managed Output Functions

Feature Description

ToDoOrchestrator is a high-level orchestration layer designed to transform high-level "vibe" prompts into structured, multi-level task trees. It leverages Output Functions as a standardized post-processing step. These functions act as the final cycle of an agent's operation, responsible for:

• Transforming raw reasoning into structured Task objects within a ToDoList.
• Enforcing hierarchical depth (LVL 1-3) based on user configuration.
• Managing Handoffs between different agent personas (e.g., from an Architect to a Developer) via context passing.
• Implementing a "Checkpoint" mechanism that halts execution to allow for user feedback/modification before any task implementation begins.

Motivation

Currently, autonomous agents often move directly from planning to execution without a reliable validation gate, leading to "hallucinated" workflows or misaligned goals. Furthermore, agent outputs frequently require fragile, ad-hoc post-processing to be useful in downstream code.

ToDoOrchestrator addresses this by:

• Providing a Human-in-the-loop (HITL) interface: Users can approve or refine the proposed plan, ensuring the agent doesn't start on the wrong foot.
• Standardizing Orchestration: By using output functions to handle the "Pyramid" decomposition, we eliminate inconsistent behavior and ensure that Level 2 tasks always inherit the correct context and dependencies from Level 1.
• Enabling Spec-Driven Development: It allows for a "vibe coding" approach where a vague idea is systematically refined into a concrete, executable specification.

Additional Context & Workflow

The orchestration follows a strict "Pyramid of Decomposition" governed by a depth_tasks configuration:

Level 0 (User): initial concept idea wich needs tasks
Level 1 (Role 1): e.g. High-level strategy and module definition.
Level 2 (Role 2): e.g. Breaking modules into technical components and dependencies.
Level 3 (Role 3): e.g. Granular, atomic implementation steps.

The Workflow Steps:

1. Input: User provides an initial prompt (the "vibe").
2. Plan Generation: The agent generates a ToDoList.
3. Output Function Execution: The system formats the tasks and triggers a Follow-up Prompt asking for user approval or modifications.
4. Parsing & Refinement: An output function parses the user's response. If changes are requested, it updates task descriptions; if approved, it triggers the next level of decomposition or starts execution.
5. Contextual Handoff: If depth > 1, the output function ensures the role-specific initial prompts and dependencies are correctly passed to the next sub-task level.