Configuration Guide
Relevant source files
The following files were used as context for generating this wiki page:
The Configuration Guide defines the structural and behavioral requirements for planning and executing technical tasks within the OhMyOpenCode ecosystem. It centers on the Prometheus planning agent, which orchestrates the transition from user requirements to a concrete, verifiable work plan. This system ensures that all implementation details, test strategies, and delegation rules are predefined before any code modifications occur.
The system enforces a strict separation between Planning (Prometheus) and Execution (Sisyphus/Atlas). Planning involves rigorous identity constraints, mandatory interview phases, and the generation of structured markdown plans that serve as the single source of truth for downstream agents.
Sources: src/agents/prometheus/identity-constraints.ts:11-30, src/agents/prometheus/behavioral-summary.ts:25-35
Planning Configuration and Identity
Prometheus operates under absolute identity constraints to prevent it from performing implementation tasks. It is strictly a planner and consultant, forbidden from writing source code or running implementation commands.
Identity Constraints
| Requirement | Detail |
|---|---|
| Role | Strategic Consultant / Plan Designer |
| Forbidden Actions | Writing source code (.ts, .py, etc.), Running implementation commands |
| Allowed Outputs | Clarifying questions, Research, Work plans (.sisyphus/plans/*.md) |
| Write Protocol | Uses atomic writes; does not append without specific tools |
Sources: src/agents/prometheus/identity-constraints.ts:32-65, src/agents/prometheus/identity-constraints.ts:147-175
Plan Generation Workflow
The transition from interview to plan generation is governed by a Self-Clearance Check. All criteria must be met before Prometheus generates the final plan.
The diagram shows the logical flow from initial request to the final execution phase.
Sources: src/agents/prometheus/identity-constraints.ts:107-118, src/agents/prometheus/plan-generation.ts:15-35
Plan Structure and Templates
Every work plan generated must follow a standardized Markdown template located at .sisyphus/plans/{name}.md. This ensures consistency for the Atlas orchestrator.
Core Plan Components
- TL;DR: Summary of objective, deliverables, and estimated effort.
- Context: Original request, interview findings, and research insights.
- Work Objectives: Definition of Done (DoD) and "Must Not Have" guardrails.
- Verification Strategy: Mandatory agent-executed QA scenarios.
- TODOs: Atomic tasks including recommended agent categories and skills.
Sources: src/agents/prometheus/plan-template.ts:11-60
Verification Configuration
A critical configuration rule is the Zero Human Intervention principle. All verification must be executable by the agent using tools like Playwright, Bash (curl), or interactive terminals.
| Deliverable Type | Verification Tool | Evidence Requirement |
|---|---|---|
| Frontend/UI | Playwright | DOM assertions, Screenshots |
| TUI/CLI | interactive_bash (tmux) | Stdout validation, Exit codes |
| API/Backend | Bash (curl) | JSON parsing, Status codes |
| Library/Module | Bash (REPL) | Function call outputs |
Sources: src/agents/prometheus/plan-template.ts:70-115
Delegation and Agent Configuration
The system utilizes a Category + Skills delegation system to route tasks to specific models optimized for different domains.
Agent Categories
The following categories are used within task() calls to determine the underlying model and reasoning effort.
| Category | Recommended Model | Use Case |
|---|---|---|
visual-engineering | google/gemini-3-pro | UI/UX, animations, styling |
ultrabrain | openai/gpt-5.3-codex | Deep logic, complex architecture |
deep | openai/gpt-5.3-codex | Goal-oriented autonomous execution |
quick | anthropic/claude-haiku-4-5 | Trivial tasks, single file fixes |
writing | google/gemini-3-flash | Documentation, prose |
Sources: src/tools/delegate-task/constants.ts:114-125
Skill Loading Protocol
When delegating, Atlas must evaluate and justify the inclusion of specialized skills.
The class diagram represents the relationship between the orchestrator, task configuration, and the skills injected into subagents.
Sources: src/agents/atlas/gpt.ts:50-70, src/tools/delegate-task/constants.ts:250-280
Execution Discipline
To maintain session continuity and prevent context loss, agents must follow strict "Todo Discipline."
- Task Registration: On start,
TodoWriteorTaskCreatemust be called with atomic steps. - Status Tracking: Each step must be marked
in_progressbefore starting andcompletedimmediately after finishing. - Atomic breakdown: Tasks requiring more than two steps must be broken down.
Sources: src/agents/hephaestus.ts:14-45
Summary
Configuration within this project is not merely a set of static variables but a rigorous protocol governing agent behavior. By enforcing strict identity constraints on Prometheus for planning, standardized Plan Templates for documentation, and a model-optimized Delegation System, the architecture ensures that complex software engineering tasks are decomposed and verified without human intervention.
Sources: src/agents/prometheus/behavioral-summary.ts:38-55, src/agents/atlas/gpt.ts:140-165