An opinionated plugin and CLI that wraps existing AI coding agents (OpenCode, Codex CLI, Claude Code) to provide multi-agent orchestration with automated model routing, context injection, and hash-anchored code editing tools.
Project overview
It provides an orchestration harness that maps task categories to different LLM providers and uses a content-hash-based edit tool to prevent stale-line corruption during automated coding sessions.
Project type
AI Agent · AI Coding
Use cases
Coding & Development
Deployment
Refer to project documentation
License
License pending
Best for
Developers and AI engineers seeking an orchestration layer to manage and route multiple AI coding agents across different LLM providers.
Users who already have paid LLM API subscriptions and use compatible harnesses like OpenCode or Claude Code.
Key capabilities
Allows one-word activation to start agents and orchestrate tasks autonomously until completion.
A multi-agent system where a lead agent orchestrates category-specialized members running in parallel and communicating through dedicated tools.
Includes built-in specialized agents, such as Sisyphus, Hephaestus, and Prometheus, for planning, deep work, architecture, and research.
An edit tool that uses content hashes to validate changes line-by-line, preventing stale-line errors and corruption.
Provides runtime-injected Model Context Protocol servers for web search, official docs, and GitHub search.
Custom skills that bring their own scoped MCP servers on demand to keep context windows clean.
Supplies IDE precision for agents, including diagnostics, navigation, workspace rename, and pattern-aware code search.
Automatically maps task categories, such as visual-engineering, deep, or quick, to the appropriate AI models.
Limitations and risks
The Codex CLI Light edition lacks agent orchestration and team_* tools required by this harness.
The codebase is currently under active refactor to support multiple agent harnesses.
Anthropic historically blocked OpenCode because of this tool's usage, which may affect users relying on Anthropic models or the OpenCode harness.
Requires external LLM APIs (Anthropic, OpenAI, Kimi, or GLM) for inference and uses PostHog for optional telemetry; inference data crosses external boundaries.
Getting started
Install via npx or bunx, select platform/providers during the non-interactive or LLM agent-driven setup, and run the ultrawork command in the CLI.