UNPKG

roocommander

Version:

Bridge Claude Code skills to Roo Code with intelligent orchestration. CLI tool + Custom Mode + 60+ production-tested skills for Cloudflare, AI, Frontend development.

54 lines (40 loc) 4.52 kB
# Roo Commander V8.1: Architecture Principles ## 1. Orchestrator (`roo-commander`) Principles * **User Interface:** Acts as the primary conversational interface for the user. * **Intent Understanding:** Focuses on understanding high-level user goals and mapping them to appropriate Manager workflows. * **Session Management:** Responsible for initiating, logging, and managing the lifecycle of work sessions. * **High-Level Delegation:** Delegates complex tasks to Manager modes via MDTM. * **Minimal Core Knowledge:** Its own `roleDefinition` and `customInstructions` are concise, primarily directing it to its `kb/` for available Manager workflows and session procedures. ## 2. Manager Mode Principles * **Domain Expertise:** Each Manager is an expert in a specific, complex workflow (e.g., project management, data product design). * **MDTM-Driven Orchestration:** Receives a high-level MDTM task from the Orchestrator and breaks it down into sub-tasks for its Squad. * **Squad Coordination:** Manages the flow of work and artifacts between its Squad members. * **Progress Tracking:** Updates its own primary MDTM task to reflect overall progress and reports status to the Orchestrator. * **KB-Guided Procedures:** Its core orchestration logic (how it sequences tasks and coordinates its squad) is defined in its `kb/` (e.g., `procedures/main-orchestration-flow.md`). ## 3. Squad Member Mode Principles * **Specialized Execution:** Highly focused on performing a single, well-defined task or producing a specific type of artifact. * **MDTM Task-Driven:** Its entire operation is guided by the detailed MDTM task assigned by its Manager. The checklist in this task is its primary instruction set. * **Artifact Producer:** The main output is typically a structured artifact. * **Self-Updating MDTM:** Updates its own assigned MDTM task to reflect progress and reports completion to its Manager. * **Dynamic KB Lookup:** Leverages its `kb/` for detailed procedures, skills, and reference material, guided by its `roleDefinition` and `kb/README.md`. ## 4. MDTM (Markdown-Driven Task Management) Principles * **Standardized Communication:** All task delegation and progress reporting between modes occurs via MDTM files. * **Hierarchical Structure:** Supports parent-child relationships (`parent_task_id`, `coordinator`) for clear work breakdown. * **Traceability:** Each MDTM task file provides a persistent, auditable record of work, including logs, status, and links to artifacts. * **Clear Contracts:** Defines objective, inputs, expected outputs, and acceptance criteria for each delegated unit of work. ## 5. Layered Knowledge Base (KB) Principles This is the cornerstone of token efficiency and effective context management: * **Tier 1: Universal Rules (`.roo/rules/`)** * **Purpose:** For fundamental, overarching principles, standards, and rules that apply to *all* modes (Roo Code default and custom). * **Content:** High-level, concise, and universally applicable (e.g., TOML+MD format, MDTM standard, session management standard). * **Loading:** Pre-loaded by Roo Code for every mode. Keep this folder lean to avoid context flooding. * **Tier 2: Mode-Specific Pre-loaded Rules (`.roo/rules-[mode_slug]/`)** * **Purpose:** For fundamental, always-relevant operational principles and context specific to a *single mode* that you want the AI to have immediately available. * **Content:** Core principles, essential operational guidelines, or very frequently used reference material for that specific mode. * **Loading:** Pre-loaded by Roo Code only when that specific mode is active. * **Structure:** This is where the `kb/` folder for each mode will reside, containing its detailed knowledge. * **Tier 3: Dynamic KB Lookup (`.roo/rules-[mode_slug]/kb/`)** * **Purpose:** For detailed, mode-specific knowledge, procedures, examples, and reference material that the AI retrieves *on demand* using its file-reading tools. * **Content:** Comprehensive procedures, specific skills, detailed examples, extensive reference data, wisdom. * **Loading:** Not pre-loaded. The AI is instructed (via its `roleDefinition` or `customInstructions`) to dynamically access and interpret this content from its `kb/` folder (which is now located within its `rules-[mode_slug]` directory), often using a `README.md` as an index. * **Benefit:** Optimizes token usage by only loading relevant context when needed.