first-fluke/oh-my-agent

What it solves

oh-my-agent provides a portable, role-based multi-agent harness that prevents AI assistants from becoming confused by splitting complex software engineering tasks across a team of specialized agents (e.g., frontend, backend, QA, PM). It eliminates the need to be locked into a single AI IDE by maintaining a single source of truth for agent skills and rules that can be projected into various runtimes.

How it works

The system uses a .agents/ directory as the central repository for skills, workflows, and rules. It projects these into the native layouts of supported AI IDEs (like Claude Code, Cursor, and Codex). Users can interact with the team via natural language chat or specific slash commands (e.g., /plan, /work, /orchestrate) to trigger structured workflows. It also features keyword-detection hooks in 11 languages to auto-activate the correct agent based on the user's prompt.

Who it’s for

Software developers and engineering teams who use AI-powered IDEs and want a structured, multi-agent approach to building features, managing infrastructure, and ensuring code quality without being tied to a specific tool.

Highlights

• Portable SSOT: Maintains agent configurations in .agents/ and emits them to open-standard artifacts, ensuring compatibility across different AI tools.
• Role-Based Specialization: Includes a vast library of pre-defined agents for specific roles like oma-architecture, oma-deepsec, and oma-tf-infra.
• Quality Gates: Implements deterministic check batteries (oma verify), independent verification workflows (ralph), and token budget caps to ensure high-quality output.
• Multi-Vendor Support: Allows mixing different AI models (e.g., Claude, GPT, Qwen) on a per-agent basis via oma-config.yaml.
• Token Efficiency: Employs a two-layer skill design to reduce token consumption by approximately 75%.