Usage Workflows

Relevant source files

This document covers common usage patterns and end-to-end workflows for different development scenarios in oh-my-openagent. It explains how agents are invoked, how work is delegated, and how the system ensures task completion through specialized orchestration layers.

For detailed information about individual agents and their specializations, see Agents. For configuration of workflow behavior, see Configuration Reference. For the underlying architecture of agent orchestration, see Agent Orchestration.

Overview

oh-my-openagent provides five primary workflow patterns, each optimized for different development scenarios:

Workflow	Primary Agent	Entry Point	Use Case
Ultrawork Mode	Sisyphus	`ultrawork` or `ulw`	Aggressive parallel execution until task completion docs/guide/overview.md136-139
Planning Workflow	Prometheus	`@plan` or Tab	Strategic planning and interview-mode requirement gathering docs/guide/overview.md106-111
Deep Work	Hephaestus	`task(category: "deep")`	Autonomous exploration and implementation using GPT-5.5 docs/guide/overview.md91-98
Consultation	Oracle, Explore	`@oracle`, `@explore`	Architecture decisions and codebase search docs/guide/overview.md118-129
CI/Headless	CLI Runner	`oh-my-opencode run`	Automated/Non-interactive task execution docs/guide/installation.md38

All workflows share common infrastructure:

BackgroundManager: Manages task lifecycle, states (pending → running), and concurrency limits docs/reference/configuration.md127-138
Intent Gate: Handles the initial classification of requests to route to the correct specialized agent docs/guide/overview.md57
Category System: Routes tasks by intent (e.g., visual-engineering, ultrabrain, deep) to the most appropriate model docs/guide/overview.md69

Sources: docs/guide/overview.md48-72 docs/guide/orchestration.md7-14 docs/reference/configuration.md127-138

Workflow Execution Model

The following diagram bridges the user's natural language intent to the underlying code entities and orchestration logic.

Diagram: Workflow Execution Flow with Code Entities

Sources: docs/guide/overview.md48-129 docs/guide/orchestration.md34-76 docs/reference/features.md3-33

Sub-Pages

9.1 Ultrawork Mode

Using /ultrawork (or ulw) to engage all agents until task completion with aggressive parallel execution. Sisyphus acts as the primary orchestrator, typically preferring high-performance models like claude-opus-4-7 or kimi-k2.6 docs/guide/overview.md83-87 This mode activates persistence mechanisms to ensure the task is driven to completion docs/guide/overview.md136-140

For details, see Ultrawork Mode. Sources: docs/guide/overview.md134-141 docs/reference/features.md13

9.2 Planning Workflow

Using /start-work with Prometheus for strategic planning. This workflow involves Prometheus interviewing the user to identify scope and ambiguities before Atlas executes the plan systematically docs/guide/orchestration.md106-111 Plans are generated as markdown files in the .omo/plans/ directory docs/guide/orchestration.md58

For details, see Planning Workflow. Sources: docs/guide/overview.md106-117 docs/guide/orchestration.md106-142

9.3 Deep Work with Hephaestus

Delegating autonomous deep work tasks to Hephaestus. Hephaestus is an autonomous GPT-native worker specialized for end-to-end execution, requiring a GPT-capable provider like gpt-5.5 docs/reference/features.md14 He works independently for extended periods to solve hard technical problems docs/guide/agent-model-matching.md58-67

For details, see Deep Work with Hephaestus. Sources: docs/guide/overview.md91-98 docs/reference/features.md14 docs/guide/agent-model-matching.md58-67

9.4 Debugging and Consultation

Using Oracle for architecture decisions, Explore for codebase search, and Librarian for external docs. These specialized agents provide high-IQ consultation and evidence-based answers docs/reference/features.md15-17 They are read-only and cannot modify the codebase directly docs/reference/features.md48-50

For details, see Debugging and Consultation. Sources: docs/guide/overview.md118-129 docs/reference/features.md15-17 docs/reference/features.md48-50

9.5 Non-Interactive and CI Mode

Using oh-my-opencode run for automated/headless workflows. This utilizes the CLI's non-interactive mode for task execution in CI pipelines docs/guide/installation.md38 It supports automated environment setup and headless agent execution docs/guide/installation.md129-131

For details, see Non-Interactive and CI Mode. Sources: docs/guide/installation.md38 docs/guide/installation.md129-131

Model Resolution in Workflows

Every workflow relies on a multi-tier model resolution system to ensure the right "brain" is used for the task, matching agent prompts to model personalities (e.g., mechanics-driven for Claude vs. principle-driven for GPT).

Diagram: Model Resolution Waterfall

Sources: docs/guide/agent-model-matching.md45-56 docs/reference/configuration.md79-97 docs/guide/orchestration.md78