🎯 RooForge

A structured multi-agent orchestration system for Roo Code

A hierarchical pipeline of specialized AI modes — from strategic planning to atomic execution.

Jump to: Overview · Pipeline · Modes · Commands · Installation · MCP Servers · Contributing

⚠️ Prerequisites

Before installing, enable the Run Slash Command experimental feature in Roo Code — the entire pipeline depends on it:

Open Roo Code settings (gear icon)
Go to Experimental Settings
Enable "Run Slash Command"
Restart VS Code if prompted

Without this setting, agents cannot execute /plan, /delegate, /blueprint, or any other slash command. See run_slash_command docs for details.

📋 Overview

This project provides a curated set of custom mode export files, slash commands, and a Forge skill that together define a disciplined, multi-layered agent orchestration workflow for Roo Code. Each mode is a specialist with a clearly defined role, connected by standardized commands that cascade into each other to eliminate duplication.

🔄 The Pipeline

flowchart TD
    User["👤 User Request"] --> O
    O["🎯 Orchestrator"] -->|"/research"| A
    A["🔍 Ask<br/><i>/web → search + read</i>"] -->|"State of Intel"| O
    O -->|"/plan"| AR
    AR["🏗️ Architect<br/><i>/clarify → /blueprint</i>"] -->|"Phased Blueprint"| O
    O -->|"/execute"| SO
    SO["⚙️ Subtask Orchestrator"] -->|"/delegate"| C
    SO -->|"/delegate"| D["🪲 Debug"]
    SO -->|"/delegate"| G
    C["💻 Code<br/><i>Implementation</i>"] -->|"result"| SO
    C -->|"on error → /debug"| D
    D --> SO
    G["📦 Git<br/><i>/git → commit</i>"] --> SO
    SO -->|"/memory"| M["💾 .memory/"]
    SO -->|"task result"| O
    O -->|"/finalize"| User2["👤 User<br/><i>Final result</i>"]

    style O fill:#4A90D9,color:#fff,stroke:#2C5F8A
    style A fill:#7B68EE,color:#fff,stroke:#4B3F8A
    style AR fill:#E67E22,color:#fff,stroke:#A05A15
    style SO fill:#27AE60,color:#fff,stroke:#1A7A42
    style C fill:#8E44AD,color:#fff,stroke:#5B2D6E
    style D fill:#C0392B,color:#fff,stroke:#8A2520
    style G fill:#F39C12,color:#fff,stroke:#B8750E
    style M fill:#2C3E50,color:#fff,stroke:#1A252F
    style User fill:#95A5A6,color:#fff,stroke:#7F8C8D
    style User2 fill:#95A5A6,color:#fff,stroke:#7F8C8D

Pipeline Phases

Phase	Mode	Purpose
1 - Intel	Ask	Eliminate unknowns via web research & codebase analysis
2 - Design	Architect	Produce a phased Blueprint with individual tasks
3 - Execute	Orchestrator	Navigate Blueprint phases, delegate tasks sequentially
4 - Atomize	Subtask Orchestrator	Break tasks into atomic subtasks
5 - Implement	Code / Debug	Write or fix code
6 - Commit	Git	Validate, stage, and commit with conventional messages

Working Memory

All modes share .memory/ as working memory — gitignored, local only. Read via codebase_search, write via /memory.

File Pattern	Purpose	Behavior
`.memory/phase-{N}-{name}.md`	One file per Blueprint phase	Append only — never duplicate
`.memory/research-{topic}-{date}.md`	One file per research run (ask mode)	New file per topic
`.memory/blocker-{desc}.md`	One file per blocker	Standalone, resolvable
`.memory/memory.md`	General fallback (no phase context)	Append only
`.memory/blueprint-{date}.md`	Auto-created by `/blueprint`	Overwrite if same date

🤖 Modes

Mode	File	Description
Orchestrator	`agents/orchestrator-export.yaml`	Strategic entry point. Performs high-level task bounding, enforces the pipeline, and delegates to specialized modes.
Ask	`agents/ask-export.yaml`	Intelligence specialist. Performs web research, codebase analysis, and generates "State of Intel" reports.
Architect	`agents/architect-export.yaml`	Technical leader. Creates detailed blueprints, system designs, and structured plans from gathered intelligence.
Subtask Orchestrator	`agents/subtask-orchestrator-export.yaml`	Execution manager. Decomposes tasks into the smallest atomic units and delegates to implementation specialists.
Code	`agents/code-export.yaml`	Implementation specialist. Writes, modifies, and refactors code. Delegates errors to Debug mode.
Git	`agents/git-export.yaml`	Version control specialist. Handles conventional commits, branch management, and repository integrity.

⚡ Slash Commands

Standardized tool call formats that cascade into each other, eliminating duplication across agent files.

Base Commands

Command	Purpose
`/complete`	`attempt_completion` format — run when work is done
`/delegate`	`new_task` format — run before delegating to any mode

Flow Commands

Command	Purpose	Used By
`/clarify`	User clarification via `ask_followup_question`	Architect
`/blueprint`	Phased planning methodology — phases with individual tasks	Architect
`/finalize`	Human-readable final output	Orchestrator

Tool Commands

Command	Purpose	Used By
`/web`	Web search + URL reader via SearXNG MCP	Ask
`/git`	Git operations (MCP-first, CLI fallback)	Git

Delegation Commands (cascade to `/delegate`)

Command	Target Mode	Purpose
`/research`	`ask`	Intel gathering
`/plan`	`architect`	Blueprint creation
`/execute`	`subtask-orchestrator`	Phase-based task execution
`/debug`	`debug`	Error resolution
`/memory`	self (direct edit)	Phase-based memory persistence
`/forge-init`	`code`	Project initialization

See the Slash Commands section above for the command tables and cascading architecture.

🧠 Forge Skill + Caveman

All modes load two skills on startup:

skills/forge/SKILL.md — Pipeline orientation: flow, command registry, mode roles, conventions
skills/caveman/SKILL.md — Token-efficient communication (auto-loaded by forge skill, full intensity)

Additional skill: skills/planning-and-task-breakdown/SKILL.md — Loaded by architect during /blueprint for structured planning methodology.

🧩 Mode Interaction Flow

flowchart TD
    U["👤 User"] -->|"Submit request"| O1["🎯 Orchestrator"]
    O1 -->|"/forge-init"| INIT["💻 Code<br/><i>Init workspace</i>"]
    INIT -->|"initialized"| O1
    O1 -->|"/research → /delegate"| A["🔍 Ask<br/><i>/web for search + read</i>"]
    A -->|"State of Intel"| O2["🎯 Orchestrator<br/><i>Assemble Master Context</i>"]
    O2 -->|"/plan → /delegate"| AR["🏗️ Architect<br/><i>/clarify → /blueprint</i>"]
    AR -->|"Phased Blueprint"| O3["🎯 Orchestrator"]
    O3 -->|"/execute → /delegate"| SO["⚙️ Subtask Orchestrator"]
    SO -->|"/delegate"| C["💻 Code / Debug"]
    C -->|"result"| SO
    SO -->|"/memory"| M["💾 .memory/"]
    SO -->|"/delegate"| G["📦 Git<br/><i>/git for commit</i>"]
    G -->|"committed"| SO
    SO -->|"task result"| O4["🎯 Orchestrator<br/><i>Evaluate & next task</i>"]
    O4 -->|"/finalize"| U2["👤 User<br/><i>Final result</i>"]

    style U fill:#95A5A6,color:#fff,stroke:#7F8C8D
    style U2 fill:#95A5A6,color:#fff,stroke:#7F8C8D
    style O1 fill:#4A90D9,color:#fff,stroke:#2C5F8A
    style O2 fill:#4A90D9,color:#fff,stroke:#2C5F8A
    style O3 fill:#4A90D9,color:#fff,stroke:#2C5F8A
    style O4 fill:#4A90D9,color:#fff,stroke:#2C5F8A
    style INIT fill:#8E44AD,color:#fff,stroke:#5B2D6E
    style A fill:#7B68EE,color:#fff,stroke:#4B3F8A
    style AR fill:#E67E22,color:#fff,stroke:#A05A15
    style SO fill:#27AE60,color:#fff,stroke:#1A7A42
    style C fill:#8E44AD,color:#fff,stroke:#5B2D6E
    style G fill:#F39C12,color:#fff,stroke:#B8750E
    style M fill:#2C3E50,color:#fff,stroke:#1A252F

🚀 Installation

Install / Update (CLI)

git clone https://github.com/weselben/RooForge.git
cd RooForge
mkdir -p ~/.roo/commands ~/.roo/skills
cp -rf commands/* ~/.roo/commands/
rm -rf ~/.roo/skills/caveman ~/.roo/skills/forge ~/.roo/skills/planning-and-task-breakdown
cp -rf skills/* ~/.roo/skills/

To install a specific version, clone by tag instead:

git clone --branch v1.2.3 --depth 1 https://github.com/weselben/RooForge.git
cd RooForge
mkdir -p ~/.roo/commands ~/.roo/skills
cp -rf commands/* ~/.roo/commands/
rm -rf ~/.roo/skills/caveman ~/.roo/skills/forge ~/.roo/skills/planning-and-task-breakdown
cp -rf skills/* ~/.roo/skills/

Why rm -rf before copy? If you previously installed a skill via npx skills add (e.g. caveman), it may exist as a single file. The repo ships skills as directories, so cp -rf would fail with "cannot overwrite non-directory with directory." Removing the old versions first ensures a clean copy. The mkdir -p ensures target directories exist on fresh installs.

See Roo Code Slash Commands docs and Skills docs for details on global directories.

Import Agent Modes

Download the export YAML files from the latest release.
Open Roo Code in VS Code.
Navigate to Roo Code Settings → Custom Modes.
Click Import and select the downloaded .yaml file(s).
The modes will appear in your mode selector.

Tip: Import all six modes for the full orchestration pipeline experience.

Configure MCP Servers

See MCP Servers below for required server setup.

🔄 Automated Releases

This repository uses automated semantic versioning powered by Conventional Commits:

flowchart LR
    P["Push to main"] --> W["GitHub Actions Workflow"]
    W --> V["Compute next version<br/>from commit messages"]
    V --> T["Create git tag<br/>(e.g. v1.2.3)"]
    T --> R["Publish GitHub Release<br/>with YAML assets"]

Commit Convention

Prefix	Version Bump	Example
`feat:`	Minor	`feat: add debug mode export`
`fix:`	Patch	`fix: correct orchestrator role definition`
`feat!:` or `BREAKING CHANGE`	Major	`feat!: redesign pipeline architecture`
`docs:`	None	`docs: update README`
`chore:`	None	`chore: update workflow`
`refactor:`	None	`refactor: simplify subtask logic`
`test:`	None	`test: add validation for exports`

🪨 Caveman — Token-Efficient Communication

Caveman enforces ultra-terse communication across the entire orchestration stack. Cuts ~65% of output tokens while keeping full technical accuracy. Auto-loaded by the Forge skill on startup** — installed as part of the skills directory (see Installation step 3).

Manual install (if not using the skills directory):

npx skills add JuliusBrussee/caveman
# Select "Roo Code" when prompted

Caveman defaults to full intensity. Switch levels anytime: "caveman ultra", "caveman lite", "stop caveman".

🔌 MCP Servers

The orchestration pipeline requires two MCP (Model Context Protocol) servers for full functionality. These servers extend the capabilities of specific modes in the pipeline.

Server	File	Required By	Purpose
SearXNG	`mcp/searxng.md`	Ask	Web search & URL reading
Git MCP	`mcp/git-mcp-server.md`	Git	Git operations (CLI fallback)

💡 See each server's documentation for full setup instructions, configuration details, and usage examples.

📁 Repository Structure

.
├── .github/
│   ├── workflows/
│   │   └── release.yml              # Auto-versioning & release workflow
│   └── ISSUE_TEMPLATE/              # Bug reports, features, questions
├── agents/
│   ├── orchestrator-export.yaml     # Orchestrator mode
│   ├── subtask-orchestrator-export.yaml  # Subtask Orchestrator mode
│   ├── architect-export.yaml        # Architect mode
│   ├── ask-export.yaml              # Ask (research) mode
│   ├── code-export.yaml             # Code (implementation) mode
│   └── git-export.yaml              # Git mode
├── commands/
│   ├── complete.md                  # /complete — attempt_completion format (includes blocked variant)
│   ├── delegate.md                  # /delegate — new_task format
│   ├── clarify.md                   # /clarify — user clarification protocol
│   ├── blueprint.md                 # /blueprint — phased planning methodology
│   ├── finalize.md                  # /finalize — human-readable output
│   ├── web.md                       # /web — web search + URL reader
│   ├── git.md                       # /git — git operations (MCP + CLI)
│   ├── research.md                  # /research — intel delegation
│   ├── plan.md                      # /plan — architectural grounding
│   ├── execute.md                   # /execute — phase-based task execution
│   ├── debug.md                     # /debug — error resolution
│   ├── memory.md                    # /memory — phase-based memory persistence
│   └── forge-init.md                # /forge-init — project initialization
├── skills/
│   ├── forge/
│   │   ├── README.md                # Forge skill overview
│   │   └── SKILL.md                 # Pipeline orientation skill
│   ├── caveman/
│   │   └── SKILL.md                 # Token-efficient communication skill
│   └── planning-and-task-breakdown/
│       └── SKILL.md                 # Planning methodology skill
├── mcp/
│   ├── searxng.md                   # SearXNG MCP server setup (Ask mode)
│   └── git-mcp-server.md            # Git MCP server setup (Git mode)
├── CONTRIBUTING.md                  # Contribution guidelines
├── LICENSE                          # Apache License 2.0
└── README.md                        # This file

🤝 Contributing

We welcome community involvement! However, please note that pull requests are not automatically accepted. All contributions go through an evaluation process.

See CONTRIBUTING.md for full details on:

Our PR evaluation process
Conventional Commits (extended) requirements
Feature branch workflow
Testing expectations

📄 License

Licensed under the Apache License 2.0.

Copyright 2026 weselben

Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at

    http://www.apache.org/licenses/LICENSE-2.0

Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.

⭐ Acknowledgments

Built for Roo Code - an AI-powered coding assistant for VS Code.
Inspired by hierarchical task decomposition and multi-agent orchestration patterns.
Caveman by JuliusBrussee - token-efficient communication skill for AI agents.

⬆ Back to top

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Repository files navigation

🎯 RooForge

⚠️ Prerequisites

📋 Overview

🔄 The Pipeline

Pipeline Phases

Working Memory

🤖 Modes

⚡ Slash Commands

Base Commands

Flow Commands

Tool Commands

Delegation Commands (cascade to `/delegate`)

🧠 Forge Skill + Caveman

🧩 Mode Interaction Flow

🚀 Installation

Install / Update (CLI)

Import Agent Modes

Configure MCP Servers

🔄 Automated Releases

Commit Convention

🪨 Caveman — Token-Efficient Communication

🔌 MCP Servers

📁 Repository Structure

🤝 Contributing

📄 License

⭐ Acknowledgments

About

Uh oh!

Releases 13

Contributors

Uh oh!

Name		Name	Last commit message	Last commit date
Latest commit History 34 Commits
.github		.github
agents		agents
commands		commands
mcp		mcp
skills		skills
AGENTS.md		AGENTS.md
CONTRIBUTING.md		CONTRIBUTING.md
LICENSE		LICENSE
README.md		README.md

Folders and files

Latest commit

History

Repository files navigation

🎯 RooForge

⚠️ Prerequisites

📋 Overview

🔄 The Pipeline

Pipeline Phases

Working Memory

🤖 Modes

⚡ Slash Commands

Base Commands

Flow Commands

Tool Commands

Delegation Commands (cascade to /delegate)

🧠 Forge Skill + Caveman

🧩 Mode Interaction Flow

🚀 Installation

Install / Update (CLI)

Import Agent Modes

Configure MCP Servers

🔄 Automated Releases

Commit Convention

🪨 Caveman — Token-Efficient Communication

🔌 MCP Servers

📁 Repository Structure

🤝 Contributing

📄 License

⭐ Acknowledgments

About

Topics

Resources

License

Contributing

Uh oh!

Stars

Watchers

Forks

Releases 13

Contributors

Uh oh!

Delegation Commands (cascade to `/delegate`)