GuidesGuides
Guides
Choose the right OmO workflow for each task.
Choosing a Workflow
| Situation | Use | Why |
|---|---|---|
| Small fix or direct question | Normal prompt | Lowest overhead. |
| Complex but you want autonomy | ulw / ultrawork | Sisyphus explores, delegates, implements, verifies, and continues. |
| Complex and precision matters | Prometheus @plan → /start-work | Planning decisions are made before implementation; Atlas executes the plan. |
| Deep GPT-native reasoning | Hephaestus | Autonomous deep worker for architecture/debugging/synthesis. |
| One narrow implementation unit | task(category=..., load_skills=[...]) | Focused Sisyphus-Junior execution with the right model profile. |
Ultrawork
ulw is the “just do it” path. Give the goal and, when possible, the surface that proves it works.
ulw fix the login error message and verify success/failure in the browserGood ulw prompts include:
- The user-facing behavior to change.
- The surface to verify: browser, CLI, API, tests, or docs build.
- Hard boundaries: what must not change.
Prometheus + Atlas
Use Prometheus when correctness matters more than speed.
- Stay in Sisyphus and type
@plan "...", or switch to Prometheus with Tab. - Answer questions until requirements are clear.
- Prometheus writes a plan under
.sisyphus/plans/. - Run
/start-work. - Atlas reads the plan, creates boulder state, delegates execution, and can resume later.
If /start-work says no active plan exists, create a Prometheus plan first or clear stale .sisyphus/boulder.json state.
Hephaestus vs Sisyphus
| Aspect | Hephaestus | Sisyphus + ulw |
|---|---|---|
| Model style | GPT-native autonomous deep worker. | Orchestrator that delegates by category and skill. |
| Best for | Architecture, hard debugging, cross-domain reasoning. | Most complex “please handle it” tasks. |
| Planning | Self-plans while executing. | Can use Prometheus plans or autonomous exploration. |
| Delegation | Uses Explore/Librarian heavily for context. | Uses category-based workers and specialists. |
Default recommendation: use ulw for most work; switch to Hephaestus when you specifically want GPT-5.4-style deep reasoning.
Installation as an Agent Task
The upstream README explicitly recommends letting an agent install OmO. When helping a user install it, the agent should:
- Ask which subscriptions/providers they have.
- Run
bunx oh-my-opencode install --no-tui ...with matching flags. - Guide
opencode auth loginin an interactive terminal when needed. - Run
bunx oh-my-opencode doctor. - Explain model setup and safe overrides.
Do not silently change model settings or disable hooks/MCPs unless the user asked for that.
Model Rule of Thumb
- Sisyphus works best on Claude Opus, Kimi K2.5, or GLM 5 style models.
- Hephaestus, Oracle, and Momus are GPT-friendly or GPT-native.
- Visual engineering benefits from Gemini-style visual/front-end reasoning.
- Explore and Librarian should stay fast and cheap.
- Runtime fallback and
fallback_modelsexist so failure moves to the next configured model instead of stopping the workflow.