OmO
Oh My OpenAgentv4.7.5

Sisyphus

The default orchestrator. Plans, delegates, executes, and refuses to stop until the work is verified.

Sisyphus is the main orchestrator. It reads intent, breaks the work into a todo list, fans out to specialists in parallel, and verifies completion before returning. The name comes from the myth — Sisyphus pushes the boulder every day and never stops. The agent is tuned to behave the same way: when work is unfinished, it goes back to work.

Default model

FieldValue
Defaultanthropic|github-copilot|opencode|vercel/claude-opus-4-7 (variant max)
StyleMechanics-driven — ~1,100-line Claude-optimized prompt
Extended thinking32k budget

Runtime fallback chain

claude-opus-4-7 (max)opencode-go\|vercel/kimi-k2.6kimi-for-coding/k2p5opencode\|moonshotai\|moonshotai-cn\|firmware\|ollama-cloud\|aihubmix\|vercel/kimi-k2.5openai\|github-copilot\|opencode\|vercel/gpt-5.5 (medium)zai-coding-plan\|opencode\|vercel/glm-5opencode/big-pickle.

Substitution rules: Agent-Model Matching. Kimi K2.5/2.6 are the recommended Claude alternatives; older GPT models are not.

When to invoke

Sisyphus is the default agent — you almost always already have it. Explicit cases:

  • A task has more than one step and needs parallel research + implementation.
  • The work needs verification before reporting done.
  • You typed ultrawork / ulw — Sisyphus drives the loop.

Tool surface

Full toolset including task, call_omo_agent, background_output, background_cancel, LSP, AST-grep, hashline edit, and the task system. Sisyphus delegates by category, not by model name — see Concepts → Category + Skill.

What it doesn't do

  • Run on incompatible GPT models — the no-sisyphus-gpt hook blocks them. GPT-5.4 and GPT-5.5 have dedicated prompt paths and are allowed.
  • Take over Prometheus's interview lane — @plan enters Prometheus mode instead.
  • Replace Hephaestus's autonomous deep work — for that, switch to Hephaestus explicitly.

Source Notes

Aligned with upstream docs/reference/features.md#core-agents and src/shared/model-requirements.ts. The orchestration flow is in Orchestration overview.

On this page