Configuration System
Relevant source files
The following files were used as context for generating this wiki page:
The Configuration System in OhMyOpenCode governs the behavioral parameters, operational constraints, and workflow definitions of the agentic ecosystem. It is primarily driven by structured system prompts and template-based configurations that define the identities and tool-access levels of specialized agents such as Prometheus, Sisyphus, and Hephaestus. The system ensures deterministic execution by enforcing strict rules through XML-tagged instruction blocks and markdown-based state management.
The architecture relies on a "Plan-then-Execute" philosophy where configurations define clear boundaries between planners (who cannot modify code) and implementers (who perform the work). This is reinforced by a directory-based configuration structure within the .sisyphus/ directory, which stores transient drafts and finalized work plans.
Agent Identity and Role Configuration
The system uses configuration prompts to define specific roles and "Hard Blocks" for each agent. These configurations prevent role-overlap and ensure that agents operate within their designated safety constraints.
Role Definitions and Constraints
| Agent | Primary Role | Path Constraints | Mutation Permissions |
|---|---|---|---|
| Prometheus | Strategic Planner | .sisyphus/plans/*.md, .sisyphus/drafts/*.md | Read-only for source; Write-only for Markdown |
| Sisyphus | Orchestrator | Workspace-wide (via delegation) | Directs subagents to mutate source code |
| Hephaestus | Autonomous Worker | Targeted files per task | Full CRUD on authorized source files |
Sources: [src/agents/prometheus/gpt.ts:14-22], [src/agents/prometheus/identity-constraints.ts:13-35], [src/agents/sisyphus/default.ts:135-145]
Verification Policy Configuration
The system enforces a "Zero Human Intervention" policy. Configuration files define that every task must have agent-executable QA scenarios using specific tools like Playwright (for UI), interactive_bash (for CLI), or curl (for APIs).
Sources: [src/agents/prometheus/plan-template.ts:74-85], [src/agents/prometheus/gpt.ts:250-270]
Workflow and Phase Configuration
The Configuration System defines a multi-phase lifecycle for every task, transitioning from intent classification to execution and final verification.
The diagram shows the standard progression of an agent's operational cycle as defined in the system prompts.
Sources: [src/agents/prometheus/gpt.ts:71-150], [src/agents/prometheus/plan-generation.ts:8-25]
Intent Tiers
Configurations categorize user requests into tiers to determine the depth of the planning process:
- Trivial: Skip heavy interview; direct action.
- Standard: Full interview + Metis review.
- Architecture: Deep interview + mandatory Oracle consultation.
Sources:[src/agents/prometheus/gpt.ts:75-81],[src/agents/hephaestus/gpt.ts:106-115]
Workspace State Configuration (.sisyphus)
The system manages state through a specialized directory structure. Agents are configured to interact with these paths specifically to maintain context across sessions.
Directory Schema
.sisyphus/plans/: Contains finalized work plans in Markdown. These are the single source of truth for execution..sisyphus/drafts/: Stores intermediate interview findings and technical decisions..sisyphus/evidence/: Stores screenshots, terminal logs, and output from QA scenarios..sisyphus/notepads/: Stores orchestrator-level learnings and issues.
Sources: [src/agents/prometheus/identity-constraints.ts:110-125], [src/agents/prometheus/plan-template.ts:18-30], [src/agents/atlas/gpt-workflow.ts:105-115]
Incremental Write Protocol
To handle large configurations (plans with 50+ tasks) that exceed token limits, the system uses an incremental write configuration:
- Skeleton Write: Initialize the file with headers and verification strategy.
- Edit-Append: Batch-insert individual tasks (2-4 at a time).
- Finalize: Read the file to verify structural integrity.
Sources:[src/agents/prometheus/gpt.ts:182-192],[src/agents/prometheus/identity-constraints.ts:188-215]
Execution Strategy and Parallelism
Configurations define how work is distributed across waves to maximize throughput.
Dependency Matrix
The configuration system mandates a dependency matrix in every plan to ensure that "Level 0" (foundations like types/configs) are completed before "Level 2" (business logic).
Sources: [src/agents/prometheus/plan-template.ts:135-155], [src/agents/hephaestus/gpt.ts:250-260]
Agent Dispatch Configuration
Agents are configured with "Recommended Profiles" for specific categories of work:
- visual-engineering: For UI/UX tasks.
- ultrabrain: For complex algorithmic logic.
- quick: For small, low-risk changes.
- git: For history management and atomic commits.
Sources:[src/agents/prometheus/plan-template.ts:180-195],[src/agents/hephaestus/gpt.ts:275-290]
Conclusion
The OhMyOpenCode Configuration System is a robust framework that transforms high-level user intent into deterministic, verifiable work units. By leveraging strictly defined agent roles, directory-based state management, and detailed plan templates, it ensures that even complex architectural changes are planned thoroughly before a single line of code is modified. This "Configuration-as-Instruction" approach provides a reliable bridge between strategic planning and autonomous execution.