Skip to content
/

Custom-Modes-Roo-Code (jtgsystems)

custom-modes-roo-code · jtgsystems/Custom-Modes-Roo-Code · ★ 173 · last commit 2026-05-14

225-agent configuration library for Roo Code covering every major software development domain with 2026 security-first standards.

Best whenAgent personas should be domain-specialized experts (225 distinct roles) rather than generalists, with cross-cutting concerns (security, debugging, engineeri…
Skip ifUsing a generalist agent for domain-specific work, Ignoring 2026 security standards
vs seeds
bmad-method(persona-based behavioral framework) but scaled to 225 agents without an integrated development workflow. The Python inj…
Primitive shape 225 total
Subagents 225
00

Summary

Custom-Modes-Roo-Code — Summary

Custom-Modes-Roo-Code is a comprehensive library of 225 specialized AI agent configurations for the Roo Code VS Code extension, organized into 10 domain categories (AI/ML, Business, Core Development, Infrastructure/DevOps, Language Specialists, Legal/Compliance, Meta-Orchestration, Security/Quality, Specialized Domains, SOTA Personas) plus 11 SOTA 2026 reasoning personas. Each agent is defined as a YAML file with slug, name, roleDefinition, customInstructions, and groups (tool permissions), then compiled into a single .roomodes file (2.6 MB) and a custom_modes.yaml using Python conversion scripts. The library emphasizes "2026 security standards" (OWASP, Zero-Trust), supports React 19+, Node.js 22+, Python 3.13+, TypeScript 5.7+, and includes agents for every major development domain. The SOTA 2026 Personas implement advanced cognitive reasoning patterns (fractal elaboration, adversarial debate, parallel synthesis).

differs_from_seeds: Most similar to BMAD-METHOD (Archetype 1 — skills-only behavioral framework with named persona files), but dramatically scaled up: BMAD ships 6 named personas, this library ships 225. Where BMAD's personas are integrated into a development workflow, these agents are domain-specialist configurations without a workflow framework — there is no Constitution → Spec → Plan → Execute structure. The closest analog in the seed corpus is BMAD's analyst.md, architect.md, dev.md etc. expanded to cover legal, fintech, gaming, blockchain, SEO, and 11 SOTA reasoning tiers. Unlike any seed, this is primarily a agent configuration library, not a development methodology.

01

Overview

Custom-Modes-Roo-Code — Overview

Origin

Created by jtgsystems (GitHub). Active development with frequent updates (last commit May 2026). Version 2026.1. Described as "Professional AI Agent Configuration Library for Roo Code — 2026 Edition."

Philosophy

"A comprehensive collection of 225 specialized AI agents designed for modern software development, following 2026 security-first principles and best practices."

Core design principles:

  • Security-First Architecture: Zero-trust, OWASP compliance, encrypted secrets
  • Performance: Sub-100ms targets, Core Web Vitals optimization
  • Type Safety: TypeScript strict mode, comprehensive validation
  • Testing: >95% coverage with unit, integration, E2E tests
  • AI Integration: LLM capabilities, vector databases, modern ML
  • Cloud-Native: Kubernetes deployment, container-first architecture

2026 Standards Framing

All agents include a "2026 Standards Compliance" section in their customInstructions. This is a consistent boilerplate that references:

  • React 19+, Node.js 22+, Python 3.13+, TypeScript 5.7+
  • Go 1.24+, Java 24, Next.js 16, .NET 9
  • Kubernetes 1.32, Terraform 1.11

SOTA 2026 Personas

Eleven advanced cognitive personas:

Tier 1 — Foundational Reasoning:

  • Core Reasoning Architect (L1 Root) — "Immutable reasoning foundation, RSC guard"
  • Formula Cascade Oracle (L2) — "Fractal formula notation master"
  • Fractal Elaborator (L3-L6) — "Infinite zoom specialist"

Tier 2 — Engineering Excellence and beyond (includes adversarial debate, parallel synthesis, etc.)

Agent Design Philosophy

Each agent file encodes:

  1. A specific professional role (e.g., "Senior Python FastAPI Engineer")
  2. 2026 technology stack context
  3. Domain-specific checklists and quality criteria
  4. Orchestration patterns and communication protocols
  5. MCP tool suite expectations

The agents are designed to be selected by a meta-orchestrator (included as one of the 225 agents: "Agent Organizer Elite") rather than used in isolation.

02

Architecture

Custom-Modes-Roo-Code — Architecture

Distribution

Standalone repository. Clone and use Python conversion scripts to generate Roo Code compatible files.

Install

git clone https://github.com/jtgsystems/Custom-Modes-Roo-Code.git

# Convert YAML agents to VS Code format
cd vs-code/
python3 convert_modes.py

# Or generate the .roomodes file directly
python3 scripts/generate_roomodes.py

Directory Tree

Custom-Modes-Roo-Code/
├── agents/                         # 225 YAML agent definitions
│   ├── ai-ml/                      # 14 agents
│   ├── business-product/           # 18 agents
│   ├── core-development/           # 58 agents (fullstack, backend, frontend, etc.)
│   │   ├── architecture/
│   │   ├── backend/
│   │   ├── frontend/
│   │   ├── fullstack/
│   │   └── general/
│   ├── infrastructure-devops/      # 25 agents
│   ├── language-specialists/       # 23 agents
│   ├── legal-compliance/           # 16 agents
│   ├── meta-orchestration/         # 37 agents
│   ├── security-quality/           # 25 agents
│   ├── sota-personas/              # 11 SOTA 2026 reasoning personas
│   │   ├── cognitive/
│   │   ├── engineering/
│   │   ├── quality/
│   │   └── reasoning/
│   └── specialized-domains/        # 16 agents
├── scripts/                        # Python generation/injection scripts
│   ├── generate_roomodes.py
│   ├── compile_modes.py
│   ├── validate_custom_modes.py
│   ├── inject_engineering_protocols.py
│   ├── inject_security_playbook.py
│   ├── inject_debugging_playbook.py
│   ├── inject_github_superpowers.py
│   ├── inject_seo_ultra_prompt.py
│   ├── inject_gap_skills.py
│   ├── update_to_2026.py
│   ├── batch_roomodes.py
│   ├── migrate_all_agents.py
│   └── ... (more)
├── vs-code/
│   ├── convert_modes.py
│   └── converted_modes/
├── .roomodes                       # Compiled output (2.6 MB, all 225 agents)
├── .roomodes.00 to .roomodes.10    # Numbered partial files (for testing)
├── custom_modes.yaml               # YAML version of .roomodes
├── CLAUDE.md                       # Claude-specific guidance
├── AGENTS.md                       # Agent documentation
├── AGENT_BRIEF.md
├── SOTA_MASTER_REGISTRY.md         # SOTA persona registry
└── schemas/                        # JSON schemas for agent validation

Rule File Format

YAML per-agent format (agents/**/*.yaml):

slug: agent-organizer
name: 🎯 Agent Organizer Elite
category: meta-orchestration
subcategory: general
roleDefinition: "..."
customInstructions: "..." (multi-line markdown with 2026 standards)
groups:
  - read
  - edit
  - command
  - mcp
version: '2026.1'
lastUpdated: '2026-05-12'

Compiled .roomodes (JSON-like YAML):

customModes:
  - slug: agent-organizer
    name: 🎯 Agent Organizer Elite
    roleDefinition: "..."
    whenToUse: "..."
    groups: [read, edit, command, mcp]
    customInstructions: "..."

Target AI Tools

Primary: Roo Code VS Code extension
Secondary: Claude Code (CLAUDE.md included), any tool that reads YAML mode definitions

Required Runtime

  • Python 3 (for conversion scripts)
  • Roo Code VS Code extension
03

Components

Custom-Modes-Roo-Code — Components

Agent Categories (225 agents total)

Category Count Key Agent Examples
Core Development 58 full-stack dev, backend engineer, frontend expert, system architect, API designer, Electron expert
Meta-Orchestration 37 agent-organizer, workflow-orchestrator, context-manager, task-distributor, performance-monitor
Language Specialists 23 Python (FastAPI/Django), TypeScript/React, Rust, Go, Java (Spring Boot), C# (.NET 9), Angular, Vue.js
Infrastructure/DevOps 25 AWS/Azure/GCP cloud engineers, Kubernetes specialist, Docker expert, SRE, platform engineer
Security/Quality 25 cybersecurity expert, penetration tester, accessibility specialist, test automator
Business/Product 18 product manager, business analyst, marketing specialist, content strategist
AI/ML 14 ML engineer, AI architect, MLOps, NLP specialist, RAG evaluator, LLM integration
Legal/Compliance 16 GDPR compliance, security auditor, corporate law, IP specialist
Specialized Domains 16 fintech, gaming, blockchain/DeFi, IoT/edge, SEO, payment systems
SOTA 2026 Personas 11 core-reasoning-architect, formula-cascade-oracle, fractal-elaborator, + others

Scripts (16 Python scripts)

Script Purpose
generate_roomodes.py Compile all YAML agents into .roomodes
compile_modes.py Alternative compiler
validate_custom_modes.py Validate .roomodes structure
inject_engineering_protocols.py Inject engineering protocols into all agents
inject_security_playbook.py Inject security standards into all agents
inject_debugging_playbook.py Inject debugging procedures
inject_github_superpowers.py Inject GitHub power-user patterns
inject_seo_ultra_prompt.py Inject SEO optimization knowledge
inject_gap_skills.py Fill gaps in agent capabilities
update_to_2026.py Migrate agents to 2026 standards
batch_roomodes.py Batch processing
migrate_all_agents.py Migration utility
vs-code/convert_modes.py Convert to VS Code format

Documentation Files

File Purpose
CLAUDE.md Claude Code-specific guidance
AGENTS.md Agent documentation
AGENT_BRIEF.md Agent brief
SOTA_MASTER_REGISTRY.md SOTA persona registry

Hooks / Commands

None. No slash commands, no hooks.

05

Prompts

Custom-Modes-Roo-Code — Prompts

Prompt 1: Agent Organizer Elite (meta-orchestration)

Technique: Role definition + numbered orchestration protocol + 2026 standards boilerplate + JSON communication format.

roleDefinition: You are an Expert agent organizer specializing in multi-agent 
orchestration, team assembly, and workflow optimization. Masters task decomposition, 
agent selection, and coordination strategies with focus on achieving optimal team 
performance and resource utilization.

customInstructions: |
  ## 2026 Standards Compliance
  This agent follows 2026 best practices including:
  - Security-First: Zero-trust, OWASP compliance, encrypted secrets
  - Performance: Sub-100ms targets, Core Web Vitals optimization
  - Type Safety: TypeScript strict mode, comprehensive validation
  - Testing: >95% coverage with unit, integration, E2E tests
  
  When invoked:
  1. Query context manager for task requirements and available agents
  2. Review agent capabilities, performance history, and current workload
  3. Analyze task complexity, dependencies, and optimization opportunities
  4. Orchestrate agent teams for maximum efficiency and success
  
  ## Communication Protocol
  Organization context query:
  {
    "requesting_agent": "agent-organizer",
    "request_type": "get_organization_context",
    "payload": {
      "query": "Organization context needed: task requirements, available agents..."
    }
  }

Technique: Declarative capability catalog + structured JSON communication protocol + aspirational performance metrics (>95% accuracy, <5s response time). The JSON communication format is a novel pattern — agents are designed to communicate with each other through structured JSON payloads, suggesting a protocol-based multi-agent architecture.


Prompt 2: SOTA 2026 Persona — Context Engineering Section

Technique: Meta-cognitive instructions about how to manage context windows.

## 🔧 Context Engineering (from TerminalSkills)

### The Context Hierarchy
Structure context from most persistent to most transient:
1. **Rules Files** (CLAUDE.md, etc.) — Always loaded, project-wide
2. **Spec / Architecture Docs** — Loaded per feature/session
3. **Relevant Source Files** — Loaded per task
4. **Error Output / Test Results** — Loaded per iteration
5. **Conversation History** — Accumulates, compacts

### Packing Strategies
- **Front-load critical context**: Put the most important information first
- **Signal vs Noise**: Include only what the agent needs for THIS task
- **Progressive disclosure**: Start high-level, drill down on demand
- **Compact when drifting**: If output quality degrades, summarize and reset context

### Anti-Patterns
- Dumping entire codebases into context (noise drowns signal)
- Ignoring rules files (missed leverage point)
- Not refreshing stale context (agent works from outdated assumptions)

Technique: Meta-cognitive context management instructions injected into agent personas. This teaches agents how to manage their own context consumption — a self-referential prompting strategy that addresses token efficiency at the agent-behavior level.


Prompt 3: Quality Screening Checklist (present in all agents)

Technique: Universal quality checklist appended to every agent's instructions.

## Quality Screening Checklist
- Publish a RACI or ownership matrix covering every workstream and ensure 
  it is acknowledged by stakeholders.
- Track tasks, blockers, and due dates in the central system (issue tracker, 
  planning board) and share the updated snapshot.
- Summarize communication artefacts (decision logs, meeting notes, escalation 
  paths) for traceability.
- Report operational metrics (cycle time, throughput, SLA adherence) and 
  highlight risks or trend regressions.

Technique: Universal post-task quality protocol injected via the inject_engineering_protocols.py script across all 225 agents. This ensures consistent output quality regardless of which specialist is used.

09

Uniqueness

Custom-Modes-Roo-Code — Uniqueness

differs_from_seeds

Most similar to BMAD-METHOD (Archetype 1 — skills with named persona files), but scaled to 225 agents versus BMAD's 6, and without the integrated development workflow. Unlike BMAD (where personas are part of a methodology: analyst → architect → dev → QA), these agents are domain specialists available for independent selection. The Python injection toolchain (inject_engineering_protocols.py, inject_security_playbook.py, etc.) is a novel pattern not present in any seed — it enables cross-cutting concern injection across all 225 agents at generation time, similar to aspect-oriented programming applied to prompt engineering. The SOTA 2026 Personas represent an unusual attempt to encode advanced reasoning architectures (fractal elaboration, adversarial debate) as Roo Code modes.

Positioning

  • Category: Agent configuration library (not a development workflow framework)
  • Niche: Roo Code users who want specialized expert personas for domain-specific tasks
  • Scale: The largest single-repo agent collection in this batch at 225 agents

Observable Failure Modes

  1. Scope inflation: 225 agents creates choice paralysis — which agent to use for a given task is unclear without a routing mechanism.
  2. Aspirational multi-agent: The meta-orchestration agents describe multi-agent coordination that Roo Code doesn't support natively — users may expect automatic agent coordination that doesn't happen.
  3. 2.6 MB .roomodes file: Loading all 225 agents simultaneously may cause performance issues in Roo Code.
  4. 2026 standards staleness: The "2026 standards" boilerplate will become outdated quickly — React 20, Node 24, etc. will require updates.
  5. No workflow structure: Without a workflow framework, users must know which of 225 agents to use and when — there is no guidance for this.
  6. JSON communication protocol is fictional: The inter-agent JSON protocol described in the meta-orchestration agents cannot actually be executed by Roo Code.
04

Workflow

Custom-Modes-Roo-Code — Workflow

No Development Workflow Defined

This library does not define a development workflow. It is a configuration library of agent personas — the user selects whichever agent is appropriate for their current task.

Agent Selection Workflow (informal)

  1. User identifies the type of task (e.g., "Python FastAPI backend work")
  2. User opens Roo Code mode selector
  3. User selects the appropriate agent mode (e.g., "Python FastAPI Engineer")
  4. User interacts with the specialized agent

Meta-Orchestration Pattern (aspirational)

The meta-orchestration agents (Agent Organizer Elite, Workflow Orchestrator, Task Distributor) define an aspirational multi-agent coordination workflow:

  1. Agent Organizer receives complex task
  2. Decomposes into subtasks
  3. Maps subtasks to appropriate specialist agents
  4. Coordinates execution across agents
  5. Aggregates results

However, in practice, Roo Code does not support true multi-agent coordination — each mode is a single LLM session. The meta-orchestration agents describe how to think about coordination, not automated multi-agent execution.

Setup Flow

git clone https://github.com/jtgsystems/Custom-Modes-Roo-Code.git
cd vs-code/
python3 convert_modes.py
# Copy output to VS Code's global custom_modes.yaml or project .roomodes

Approval Gates

None defined by the framework.

Phase-to-Artifact Map

None defined. Individual agents may produce artifacts (code, documentation, test files) based on their domain expertise.

06

Memory Context

Custom-Modes-Roo-Code — Memory & Context

Memory Model

No persistent memory framework defined. This library provides agent personas, not a memory system.

Context Loading Pattern (from agent instructions)

Each agent's customInstructions includes the "Context Engineering" section (injected via Python scripts) that defines a context hierarchy:

  1. Rules Files (CLAUDE.md, etc.) — Always loaded
  2. Spec/Architecture Docs — Per feature/session
  3. Relevant Source Files — Per task
  4. Error Output/Test Results — Per iteration
  5. Conversation History — Accumulates, compacts

Agent Communication via JSON (aspirational)

The meta-orchestration agents define a JSON-based communication protocol for inter-agent context sharing:

{
  "requesting_agent": "agent-organizer",
  "request_type": "get_organization_context",
  "payload": {
    "query": "..."
  }
}

This is aspirational — Roo Code does not natively support agent-to-agent communication. This pattern documents the intended communication format if a multi-agent runtime were available.

Memory Type Summary

Dimension Value
Memory type none (session-only per Roo Code)
Persistence scope session
State files none defined
Search mechanism none
Compaction implied (from context engineering section)
07

Orchestration

Custom-Modes-Roo-Code — Orchestration

Multi-Agent

Aspirational. The meta-orchestration agents define multi-agent coordination patterns (Agent Organizer Elite, Workflow Orchestrator, Task Distributor, Multi-Agent Coordinator), but these are not backed by an actual multi-agent runtime. Each agent is a Roo Code mode — single LLM sessions.

Orchestration Pattern

Documented: hierarchical (described in agent instructions). Actual: none (single-agent execution per session).

Isolation Mechanism

None.

Multi-Model

No. Model is selected globally in Roo Code settings.

Execution Mode

Interactive loop (Roo Code extension sessions).

Consensus Mechanism

None.

The Aspirational vs Actual Gap

The meta-orchestration agents describe sophisticated patterns:

  • Sequential execution
  • Parallel processing
  • Pipeline patterns
  • Map-reduce workflows
  • Event-driven coordination
  • Hierarchical delegation
  • Consensus mechanisms
  • Failover strategies

But these are documentation for how a human should coordinate agents — they don't run automatically.

Summary Table

Dimension Value
Multi-agent no (aspirational documentation only)
Orchestration pattern none (documented: hierarchical)
Max concurrent agents 1
Isolation none
Consensus none
Prompt chaining no
Multi-model no
Execution mode interactive-loop
Crash recovery no
08

Ui Cli Surface

Custom-Modes-Roo-Code — UI / CLI Surface

Dedicated CLI Binary

None. Setup uses Python scripts.

Local UI / Dashboard

None. Uses Roo Code's built-in mode selector.

IDE Integration

Primary: Roo Code VS Code extension

Integration:

  • .roomodes file (2.6 MB) — all 225 agents compiled into Roo Code's custom modes format
  • custom_modes.yaml — YAML version
  • .roomodes.00 through .roomodes.10 — numbered partial files for testing/selective loading

After placing .roomodes in project root or copying to VS Code global config, all 225 modes appear in Roo Code's mode selector.

Secondary: Claude Code (CLAUDE.md included)

Python Toolchain

The scripts directory provides a substantial Python toolchain for agent management:

  • generate_roomodes.py — compile all agents
  • validate_custom_modes.py — validate output
  • inject_*.py — inject cross-cutting concerns (security, debugging, engineering protocols) across all agents
  • update_to_2026.py — bulk update agent references to 2026 standards

Observability

None. No logging, audit trail, or monitoring.

Cross-Tool Portability

Low. The .roomodes format is Roo Code-specific. The YAML agent definitions could be adapted for other tools with effort.

Related frameworks

same archetype · same primary tool · same memory type

BMAD-METHOD ★ 48k

Provides a full agile delivery lifecycle with named expert-persona AI collaborators that elicit the human's best thinking rather…

Agent OS ★ 4.6k

Extracts implicit codebase conventions into token-efficient markdown standards files and injects them selectively into AI agent…

Claude Conductor ★ 367

Gives Claude Code a persistent, cross-linked, auto-analyzed documentation system so it retains codebase context across sessions.

Spec-Driver (Greenfield Spec-Driven Development) ★ 25

Prevents spec rot in AI-assisted development by making implementation changes flow back into evergreen, authoritative specs via…

Anthropic Knowledge Work Plugins ★ 16k

Role-specialized plugin bundles with live MCP connectors that turn Claude into a domain expert for enterprise knowledge workers.

Codex Integration for Claude Code (skill-codex) ★ 1.3k

Single Claude Code skill that handles Codex CLI invocation correctly (stdin blocking, thinking token suppression, session resume)…