This is an early release preview. You may encounter bugs.
Oh My Opencode Slim logo
Unclaimed

Tool coding

Oh My Opencode Slim

Lean, fine tuned Opencode multi agent suite · Mix any models · Auto delegate tasks

A 83/100 GitHub score ? This grade is derived from GitHub signals, not user votes. Open for the full breakdown.
Discussion

01 / Summary

About Oh My Opencode Slim.

Oh My Opencode Slim is an agent-orchestration plugin for OpenCode, a slimmed-down fork of Oh My Opencode that keeps the core orchestration and drops the rest. It installs a team of seven specialized agents under one Orchestrator: Explorer scouts the codebase, Librarian retrieves external documentation, Oracle advises on architecture and hard debugging, Council sends one question to several models in parallel and synthesises a single answer, Designer handles UI work, and Fixer carries out scoped implementation tasks. An optional Observer reads images, screenshots, PDFs, and diagrams for an Orchestrator whose model is not multimodal.

Each agent runs on its own model, mixed across any providers OpenCode can see, so the plugin trades off quality, speed, and cost per role rather than asking one model to do everything. The Orchestrator plans the work graph, dispatches specialists as background tasks, tracks them, and reconciles their results before continuing, with parallel work as the default. Presets name the whole team's models and switch at runtime, and a preset can inherit from one base preset and override single agents. Generated presets run the team on OpenAI models or on OpenCode Go models, and the per-agent guidance recommends models from Anthropic, OpenAI, DeepSeek, Zhipu, Qwen, MiniMax, Xiaomi, Moonshot, and Google.

Eight bundled skills are prompt-based playbooks assigned per agent: codemap for hierarchical repository maps, deepwork for large multi-phase sessions with review gates, verification-planning before non-trivial changes, simplify for behavior-preserving cleanup, worktrees for isolated coding lanes, clonedeps for inspecting dependency source, reflect for turning repeated friction into reusable skills or config, and a self-configuration skill. Skill and MCP assignments are permission grants, so an agent activates only what it has been given.

Around the agents sit code-intelligence tools (LSP tools, AST-aware search across 25 languages, and the Context7 and gh_grep MCPs for documentation and GitHub code search), a multiplexer integration that opens each agent in a Tmux, Zellij, Herdr, cmux, or kitty pane, an optional floating Companion window that shows which agents are active, ACP agents such as Claude Code or Gemini as delegatable subagents, a marketplace for community agent packages, and an Interview flow that turns a rough idea into a markdown spec through a browser-based Q&A before implementation. Custom agents, prompt overrides, per-agent permissions, global disable switches, and project-local configuration cover customization, and the plugin runs on both OpenCode v1 and v2 hosts.

  • Seven specialized agents: Orchestrator, Explorer, Oracle, Council, Librarian, Designer, and Fixer, each on its own model from any provider, plus an optional multimodal Observer
  • Background orchestration: the Orchestrator dispatches specialists as parallel background tasks, tracks them, and reconciles results before continuing
  • Council: runs several models in parallel on the same question and synthesises one answer, invoked with @council
  • Preset switching: swaps the whole team's models at runtime with a slash command, with single-parent preset inheritance
  • Bundled skills: codemap, deepwork, verification-planning, simplify, worktrees, clonedeps, reflect, and a self-configuration skill, assigned per agent as permission grants
  • Code intelligence: LSP tools, AST-aware search across 25 languages, and built-in MCPs for documentation and GitHub code search
  • Multiplexer integration: opens each agent in a dedicated Tmux, Zellij, Herdr, cmux, or kitty pane to watch specialists work live
  • Companion: an optional floating desktop window showing the session state and active agents
  • ACP agents: connects external ACP-compatible agents such as Claude Code or Gemini as delegatable subagents with progress streamed to the TUI
  • Interview to spec: turns a rough idea into a structured markdown spec through a browser-based Q&A, then implements it
  • Customization: custom agents, prompt overrides, per-agent skill and MCP permissions, global disable switches, project-local configuration, and marketplace agent packages
  • OpenCode v1 and v2: the same plugin runs on both hosts, with a documented feature matrix and version pinning

02 / Discussion CREDIBILITY-GATED

Discussion

Reading is open to everyone. Posting and voting need a verified identity or a GitHub grade of B or higher.

  • No discussions yet.

04 / Build

Build with Oh My Opencode Slim.

Browse the catalog for frameworks, tools, and harnesses, each scored on real GitHub credibility.

Get Oh My Opencode Slim →

Browse the catalog