Oh My OpenCode Slim
An OpenCode plugin that routes development work across specialized, configurable multi-model agents.
- Source repo
- alvinunreal/oh-my-opencode-slim
- Stars
- ★ 10k
- Last updated
- 1d ago
- License
- MIT
- Primary language
- TypeScript
- FA score
- 50/100 · Major gaps
At a glance
- How it runs
- Works with
- Platform-specificOpenAI APIClaude Code (Partial support)
- Cost
- Free software; you pay for model usage
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- An engineer joining a large OpenCode project who wants Explorer to map the codebase before implementation begins.
- Not a fit if
- Teams not using OpenCode; the plugin cannot run standalone
- Users wanting zero-config defaults without provider auth setup
- Source review
- 50/100 · Major gaps
What does this agent do, and when should you use it?
Oh My OpenCode Slim is an agent-orchestration plugin for OpenCode, not a standalone application or general-purpose API service. Its Orchestrator plans work, dispatches specialist tasks in the background, and reconciles their results before proceeding. The built-in team consists of Orchestrator, Explorer, Oracle, Council, Librarian, Designer, and Fixer, with an optional Observer for visual files. Local OpenCode configuration assigns models, skills, and MCP permissions to individual roles, including supplied OpenAI and OpenCode Go presets. It fits developers who want a configurable, observable parallel-agent workflow inside OpenCode and are prepared to manage OpenCode setup and provider authentication.
The installer adds the plugin to the plugin array in ~/.config/opencode/opencode.json and generates ~/.config/opencode/oh-my-opencode-slim.json. Orchestrator creates a work graph and schedules background specialists; Explorer scouts a codebase, Librarian retrieves external information through websearch, context7, and gh_grep, Oracle handles architecture, difficult debugging, and review, and Designer and Fixer perform UI and bounded implementation work. Users can delegate explicitly with @agentName <task>, run @council to send one question to multiple models in parallel and synthesize an answer, and switch full model setups with /preset. The plugin also exposes LSP tools, AST-aware search across 25 languages, and prompt-based skills including codemap, deepwork, and verification-planning. When enabled, Observer uses the read tool to return structured observations from images, screenshots, PDFs, and other visual files.
- An engineer joining a large OpenCode project who wants Explorer to map the codebase before implementation begins.
- A staff engineer comparing two architecture options who wants
@councilto gather parallel model perspectives and synthesize them. - A developer investigating an unfamiliar dependency or current documentation who assigns Librarian to use its permitted research MCPs.
- A developer with several bounded code changes who wants Fixers to work in the background while the Orchestrator remains available in the session.
- A team whose main Orchestrator is not multimodal but needs to analyze screenshots or PDFs through an enabled Observer.
How do you install or deploy this agent?
Prerequisites: OpenCode and authentication for the model providers you plan to use. Install with Bun: bunx oh-my-opencode-slim@latest install. If Bun is unavailable, use: npx oh-my-opencode-slim@latest install. Then run opencode auth login and opencode models --refresh. Installation generates the OpenAI and OpenCode Go presets and activates OpenAI by default; to activate OpenCode Go at install time, run bunx oh-my-opencode-slim@latest install --preset=opencode-go.
How do you use this agent?
Start opencode, then run ping all agents to verify that configured roles respond. Edit ~/.config/opencode/oh-my-opencode-slim.json and set each role's available model, optional variant, skills, and mcps under presets. Delegate manually with @agentName <task>, invoke multi-model synthesis with @council <question>, and use /preset to change presets at runtime. For a floating activity display, install with --companion=yes; the repository also documents multiplexer viewing through Tmux, Zellij, Herdr, cmux, or kitty panes.
What are this agent's strengths and limitations?
- It separates orchestration, codebase reconnaissance, research, architectural judgment, UI work, and implementation into named specialist roles instead of presenting one undifferentiated assistant.
- Background specialist dispatch is the default orchestration model, allowing parallel work while the Orchestrator continues the session.
- Model, skill, and MCP access are configurable per agent, with supplied OpenAI and OpenCode Go presets plus runtime
/presetswitching. - Council provides a distinct multi-model parallel-and-synthesis path for decisions that benefit from competing perspectives.
- Concrete developer tooling includes LSP tools, AST-aware search across 25 languages, and the
websearch,context7, andgh_grepMCPs.
- Its core delivery boundary is OpenCode; the supplied evidence does not establish standalone-library, standalone-web-service, or native operation in other chat products.
- Adoption requires provider setup and
opencode auth login; available models and usage costs depend on the chosen providers and preset. - Council runs several models concurrently and is described as usually the system's highest-cost path, so automatic use is intentionally restricted.
- Observer is disabled by default; visual analysis requires enabling it and configuring a vision-capable model.
- Only part of the documentation is supplied, so the complete behavior of customization options, ACP configuration, and troubleshooting cannot be verified here.
How does this agent compare with similar options?
The repository supplies OpenAI and OpenCode Go presets. The OpenAI preset assigns Terra, Sol, and Luna to different roles; the OpenCode Go preset uses OpenCode Go models and enables Observer because its Orchestrator is not multimodal. These are configuration choices within the same OpenCode plugin, not separate competing products.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Oh My OpenCode Slim This agent | 50 · Major gaps | Agent plugin / skillFree + model costs | ★ 10k | 1d ago | TypeScript | OpenAI API |
| Oh My OpenAgent | 49 · Major gaps | Agent plugin / skillFree + model costs | ★ 71k | today | TypeScript | Claude Code |
| AgentScope 2.0 | 70 · Some gaps | Library / SDKFree + model costs | ★ 36k | 2d ago | Python | OpenAI API · Claude API |
| npcpy Agent Framework | 43 · Major gaps | Library / SDKFree + model costs | ★ 1.5k | 4d ago | Python | Codex · Claude Code · OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence: The plugin installer modifies global config files (~/.config/opencode/opencode.json) and generates preset configs with model and MCP permissions. Skill assignment is permission-based, with default skills assigned. Installer asks about enabling Companion, default no. Data flow: docs describe Orchestrator dispatching background agents, but no explicit data flow or sensitive data handling. Dependencies include third-party packages, but no security audit evidence. External effects: modifies config, launches background processes (Companion), clones dependencies, but no explicit user confirmation. Rollback: docs mention config reset, but no automatic rollback. Source attribution: code and docs clearly attribute to author and repo, but publisher unverified. Deductions: lack of explicit least-privilege principle, user confirmation, data flow transparency, sensitive data handling; dependency security unverified; external effects not fully documented; rollback incomplete.
Evidence: README and docs are consistent, multiple docs cross-reference, good self-consistency. Dependencies listed in package.json, but no lockfile or version pinning evidence. Failure messages: docs mention troubleshooting, but no specific error handling. Deductions: dependency availability unverified, failure message handling insufficient.
Evidence: README targets developers, provides multiple use cases (codebase investigation, doc reference, architecture review, UI work, implementation tasks). Capability boundaries defined via skills and MCP permissions. Trigger precision via @agentName and skill invocation. Environment fit: supports multiple model providers and presets, but no explicit OS/version requirements. Deductions: environment fit info incomplete.
Evidence: Clear information architecture with README, docs directory, config reference. Install notes detailed, including manual and automatic. Naming stable, version numbers clear. Examples and FAQ: README includes user testimonials and quick start, but no FAQ. Known limitations: docs mention Council cost, but no systematic list. License: MIT license clear. Versioning: package.json has version, but no CHANGELOG. Maintenance responsibility: SECURITY.md and contributor guide, but no explicit maintainer. Deductions: missing FAQ, CHANGELOG, and explicit maintainer.
Evidence: Output usability: generated configs and skills directly usable. Marginal value: multi-agent orchestration provides advantage over single model. Cost-benefit: docs mention $30 preset, but no detailed cost analysis. Deductions: cost-benefit analysis insufficient.
Evidence: README claims (features, performance) lack test or benchmark data. Cross-source corroboration: user testimonials from community, but no verifiable links. Fact-inference separation: mythological descriptions mixed with functional descriptions, but not clearly separated. Deductions: lack of verifiable evidence and fact/inference separation.
- The plugin modifies global config files and may launch background processes; back up config before installation.
- Dependencies lack security audit evidence; assess supply chain risks yourself.
- Publisher identity is unverified; use with caution.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
What's Changed * docs: overhaul Pantheon recommended models, align zh-CN hero, and add $40 dual-subscription preset by @zjm54321 in https://github.com/alvinunreal/oh-my-opencode-slim/pull/1669 * refactor(background): dissolve board ceremony — watchdog abort, structured v2 registration, gate trimming (DRAFT, rebase planned onto #1652) by @Jiajun0413 in https://github.com/alvinunreal/oh-my-opencode-slim/pull/1662 * fix(foreground-fallback): send only the internal reminder on replay by @dhaern in https://github.com/alvinunreal/oh-my-opencode-slim/pull/1671 …View on GitHub ↗ -
Release notes excerpt
What's Changed * refactor(deepwork): drop the speculative claims path; unify the tombstone flip predicate by @Jiajun0413 in https://github.com/alvinunreal/oh-my-opencode-slim/pull/1618 * fix(agents): deny question for subagents by default by @Jiajun0413 in https://github.com/alvinunreal/oh-my-opencode-slim/pull/1631 * fix(companion): stabilize multi-monitor popup and native dragging by @Aveer in https://github.com/alvinunreal/oh-my-opencode-slim/pull/1624 * fix: honor agent display names in sidebar by @giannisCKS in https://github.com/alvinunreal/oh-my-opencode-slim/pull/1619 …View on GitHub ↗ -
Release notes excerpt
What's Changed * perf(task-reply): drop the dead v2 answers arg and its caveat sentence by @Jiajun0413 in https://github.com/alvinunreal/oh-my-opencode-slim/pull/1510 * fix(v2): settle revived children after compaction by @guidan-nick in https://github.com/alvinunreal/oh-my-opencode-slim/pull/1512 * fix(foreground-fallback): walk the chain from the head and skip only tried or abandoned providers by @dhaern in https://github.com/alvinunreal/oh-my-opencode-slim/pull/1511 …View on GitHub ↗
FAQ
Is it limited to one model provider?
How are networked MCP permissions controlled?
mcps setting. The supplied example gives Librarian websearch, context7, and gh_grep, while other agents may have none.What should I do if an agent does not respond?
ping all agents in OpenCode. The README directs users to check provider authentication and the configuration file when a configured agent fails to respond.