ohmg_skill

Overview

This skill is an orchestrator for multi-agent software projects, coordinating specialized domain agents (PM, Frontend, Backend, Mobile, QA, Debug) and managing shared state through Serena Memory. It streamlines parallel work, spawns agents via CLI, and provides real-time observability with terminal and web dashboards. The tool integrates multiple LLM CLIs so you can map each agent to the vendor that best fits the task.

How this skill works

The orchestrator listens for coordination commands and triggers specialized agents to perform domain-specific tasks, using Serena Memory to write structured state files for cross-agent context and monitoring. Agents are spawned from the CLI or dashboard, run tasks in isolated sessions, and report progress and artifacts back into the shared memory directory for real-time inspection. Integration hooks map agents to different LLM CLIs, and an SSE bridge mode enables environments that require stdio-to-HTTP bridging.

When to use it

• Coordinating complex, multi-domain engineering projects
• Parallelizing work across PM, Frontend, Backend, Mobile, QA, and Debug agents
• Maintaining shared, observable state across agents with Serena Memory
• Setting up real-time dashboards for agent workflows and progress tracking
• Bridging environments that require stdio-to-HTTP (SSE) connectivity

Best practices

• Define a clear Workflow Guide prompt to coordinate multi-step runs and handoffs
• Map each domain agent to the best-suited LLM CLI in user-preferences.yaml
• Keep Serena Memory under versioned control and use structured keys for predictable monitoring
• Use the dashboard for high-level observability and the CLI for scripted agent spawning and automated runs
• Run bunx oh-my-ag doctor regularly to validate environment and dependencies

Example use cases

• Run a single coordinated project: PM decomposes tasks, frontend/backend agents implement features, QA agent validates, Debug agent investigates failures
• Parallelize feature work across multiple agents to shorten delivery time while retaining centralized state
• Integrate different LLM vendors per role (e.g., Gemini for frontend, Codex for backend, Claude for PM/QA) using the agent_cli_mapping
• Deploy a live monitoring setup: start the dashboard to watch .serena/memories/ and follow agent progress
• Bridge a CI environment that lacks direct stdio access by using SSE bridge mode