The oMO Pipeline

oh-my-openagent (oMO) uses a multi-agent pipeline to turn your intent into working code. Three agents — Prometheus, Atlas, and Sisyphus: each handle a different stage. A fourth, Oracle, is a read-only consultant you can call in when you're stuck.

The Flow

  You type /start-work "Build a bookmark API with auth"
       │
       ▼
  ┌──────────┐
  │Prometheus │  Creates a structured plan
  │ (Planner) │  Saves to .sisyphus/plans/*.md
  └────┬─────┘
       │
       ▼
  ┌──────────┐
  │   Atlas   │  Reads the plan, delegates tasks
  │(Orchestr.)│  Verifies each step
  └────┬─────┘
       │
       ▼
  ┌──────────┐
  │ Sisyphus  │  Executes tasks via category agents
  │ (Executor)│  Routes to specialists automatically
  └──────────┘
       │
       ├──► explore (search codebase)
       ├──► oracle (architecture/debugging)
       ├──► librarian (docs & references)
       └──► category agents (implementation)
              ├── visual-engineering
              ├── deep
              ├── ultrabrain
              ├── artistry
              ├── quick
              └── writing

Two Modes of Operation

The most important thing to understand: the pipeline only fires when you use /start-work. The rest of the time, you're talking directly to Sisyphus.

Pipeline mode (/start-work)

For significant new work — new projects, major features, multi-step refactors. You describe what you want, and the full pipeline runs:

1. Prometheus creates a plan
2. Atlas orchestrates execution
3. Sisyphus (and category agents) implement

Direct mode (default)

For everything else — quick fixes, questions, small changes, debugging. You talk to Sisyphus, and it handles things directly or delegates as needed.

"Add a comment to the calculateTotal function"
→ Sisyphus handles it directly

"What does this config setting do?"
→ Sisyphus answers directly

"Fix the login bug"
→ Sisyphus explores, diagnoses, fixes — no pipeline needed

Decision Tree

Is this significant new work (new project, major feature, large refactor)?
│
├── YES → Use /start-work
│          Prometheus plans, Atlas orchestrates, Sisyphus executes
│
└── NO → Talk to Sisyphus directly
         │
         ├── Need to clarify intent first?
         │   └── YES → Use /grill-me or /grill-with-docs
         │
         ├── Stuck after 2+ failed fix attempts?
         │   └── YES → Consult Oracle
         │
         └── Just do it
             └── Sisyphus routes to the right agent

The Agents

Agent	Role	Trigger	Output
Prometheus	Planner	/start-work	.sisyphus/plans/*.md
Atlas	Orchestrator	Automatic (after Prometheus)	Delegation to category agents
Sisyphus	Executor	Always (your default agent)	Code changes, answers
Oracle	Consultant	Manual or auto (2+ failures)	Analysis, recommendations

Category Agents

Sisyphus delegates implementation work to category agents based on the task type. See Category Agents for the full breakdown.

Category	Optimized for
quick	Small, fast tasks — comments, renames, one-liners
deep	Multi-step implementation with verification
ultrabrain	Complex reasoning — architecture, algorithms
visual-engineering	UI/UX with design quality enforcement
artistry	Creative work — copy, design, polish
writing	Documentation, articles, prose

Real-World Examples

See the Playbooks for step-by-step walkthroughs:

• Start a New Project — Full pipeline with /start-work
• Add a Feature — Direct Sisyphus delegation
• Debug a Bug — Sisyphus + Oracle for hard bugs
• Refactor with Grill — Clarify before you code