Agent System

Oh My OpenAgent runs a team of specialized AI agents. Each agent is tuned to a specific model’s strengths. You don’t pick models — you describe work, and the framework routes it.

Discipline Agents

Three core agents form the discipline layer. They don’t just chat — they ship.

Sisyphus — The Main Orchestrator

Model: claude-opus-4-6 / kimi-k2.5 / glm-5

Sisyphus is your main orchestrator. He plans, delegates to specialists, and drives tasks to completion with aggressive parallel execution. He does not stop halfway.

What Sisyphus does:

Analyze tasks and decompose them into subtasks
Select the best agent for each subtask by category (not by model name)
Coordinate parallel execution across multiple agents
Integrate results and drive the task to 100% completion

Hephaestus — The Autonomous Deep Worker

Model: gpt-5.4

Give him a goal, not a recipe. Hephaestus explores the codebase, researches patterns, and executes end-to-end without hand-holding. The Legitimate Craftsman.

Best for:

Large refactoring tasks where you want autonomous exploration
Research-heavy work that needs deep investigation
Tasks where you know the destination but not the path

Prometheus — The Strategic Planner

Model: claude-opus-4-6 / kimi-k2.5 / glm-5

Prometheus is your strategic planner. Activate with /start-work — it enters interview mode: questions you, identifies scope and ambiguities, builds a detailed plan before a single line of code is touched. Equipped with Metis (Plan Consultant).

When to use Prometheus:

Complex multi-step tasks where planning saves time
Tasks with unclear requirements that need clarification
When you want a verified plan before execution begins

Expert Agent Team

Sisyphus dispatches specialist agents on demand. Each has a focused domain:

Agent	Role	Best For
Oracle	System architect	Architecture design, debugging complex systems, performance analysis
Librarian	Documentation specialist	Searching docs, code search, knowledge retrieval
Explore	Codebase analyst	Deep codebase analysis, fast grep, pattern discovery
Multimodal Looker	Visual analyst	Analyzing screenshots, design mockups, visual content

These experts are dispatched automatically by Sisyphus. You don’t need to call them directly — just describe your task.

Agent Orchestration Categories

When Sisyphus delegates to a subagent, it doesn’t pick a model. It picks a category. The category maps automatically to the right model:

Category	What it’s for
`visual-engineering`	Frontend, UI/UX, design
`deep`	Autonomous research + execution
`quick`	Single-file changes, typos
`ultrabrain`	Hard logic, architecture decisions

Agent says what kind of work. Harness picks the right model. ultrabrain routes to GPT-5.4 xhigh by default. No manual configuration needed.

You can define custom categories in configuration. See Configuration.

ultrawork / ulw

One word activates everything.

ultrawork: refactor the authentication module

or simply:

ulw: fix all lint warnings

What happens when you type ultrawork:

Sisyphus analyzes your task
Decomposes into subtasks by category
Dispatches agents in parallel
Monitors progress and reassigns if needed
Integrates results and delivers

It doesn’t stop until the task is 100% done. That’s the promise.

Background Agents

Fire 5+ specialists in parallel:

Context stays lean — each agent gets only what it needs
Results return when ready
Main orchestration continues without blocking
Perfect for independent subtasks like research, testing, documentation

Background agents are activated automatically by Sisyphus during ultrawork mode. You can also configure concurrency limits in Configuration.

Claude Code Compatibility

Every hook, command, skill, MCP, and plugin from Claude Code works unchanged. Full compatibility, including plugins. No migration needed.

Next Steps

Tools — The tools agents use: Hashline, LSP, AST-Grep, MCPs
Features — Ralph Loop, Todo Enforcer, Comment Checker
Configuration — Tune agent models, categories, and behavior