Show HN: Oh-my-agent – A structural harness for AI agents in real projects
otti-sister Tuesday, March 17, 2026
If you ask an AI agent to build a todo app, it will usually produce something. But in a real development environment, limitations become obvious: hallucinations, drifting off task, or repeating the same mistakes.
Looking at current prompts and skills, a few recurring problems show up: missing critical library versions, vague personas like you are a senior engineer, and bloated prompts that burn tokens without improving outcomes.
oh-my-agent addresses this by introducing a structured protocol to keep agents aligned with real project constraints.
Key Ideas: 1. Clarification protocol: Analyzes requirement ambiguity to decide whether to proceed, offer options, or stop for clarification.
2. Difficulty guide: Adjusts protocol depth based on task complexity.
3. Context budget: Enforces token limits per model to prevent unnecessary usage.
OpenAI’s Symphony release this March reinforced the importance of this approach. Symphony is powerful for autonomous workflows, but it performs best when the codebase is designed to be AI-friendly.
OpenAI refers to this as harness engineering: structured documentation like ARCHITECTURE.md, AGENTS.md, and WORKFLOW.md that agents can reliably interpret. oh-my-agent is built around these principles. Instead of just sending prompts, it provides the structural harness agents need to understand project context and operate autonomously.
Currently supports: Antigravity, Claude Code, Codex CLI, and Cursor.
The core idea: Using AI agents effectively isn’t about clever prompts—it’s about the engineering structure around them.