Oh My Crew

A role-based agent fork of oh-my-openagent

Clean, role-descriptive agent names with content-filter-safe system prompts for AI proxy providers.

English | Bahasa Indonesia

What is this?

Oh My Crew is a fork of oh-my-openagent (by @code-yeongyu), based on oh-my-china (by @enowdev), with the following modifications:

1. System prompts adjusted to pass through AI proxy content filters
2. All agents renamed to clean, role-descriptive crew names -- no political, cultural, or mythological references

Agent Roster

| Original | Crew Name | Role | |----------|-----------|------| | Sisyphus | Captain | Main orchestrator. Steers the ship. | | Hephaestus | Strategist | Autonomous deep worker. Plans and executes. | | Oracle | Sage | Read-only consultant. Wisdom without action. | | Librarian | Scribe | External docs search. Finds written knowledge. | | Explore | Scout | Codebase search. Reconnaissance specialist. | | Atlas | Foreman | Todo orchestrator. Keeps work on track. | | Prometheus | Architect | Strategic planner. Designs before building. | | Metis | Advisor | Pre-planning consultant. Identifies risks. | | Momus | Auditor | Plan reviewer/critic. The quality gate. | | Sisyphus-Junior | Cadet | Task executor. Follows orders. | | Multimodal Looker | Lookout | Vision/PDF analysis. Sharp-eyed observer. |

Content Filter Fix

The phrase "Powerful AI Agent" combined with identity override directives triggers some AI proxy content filters. This fork replaces it with softer phrasing that passes through.

Session Guardian (NEW)

Built-in skill that gives agents autonomous session lifecycle management:

| Feature | What It Does | |---------|-------------| | Git Checkpoints | Auto-commits after each completed task. One commit per task, conventional commit messages. | | Context Monitoring | Tracks context window usage via heuristics + system signals. Knows when to stop. | | Smart Handoff Timing | Never interrupts mid-task. Finishes current work first, then creates handoff. | | Structured Handoff Docs | Generates HANDOFF_DOC/handoff-NNN.md with full context for the next session. | | Chain Continuity | Each handoff carries forward decisions and context from ALL previous sessions. | | Copy-Paste Resume | Outputs a ready-to-paste prompt for starting the next session seamlessly. | | Credentials Protection | Manages .credentials/ folder for API keys, tokens, passwords. Auto-added to .gitignore. Agents never hardcode secrets in source code. |

How it works:

Agent works -> completes task -> git commit -> checks context level
  |- Context OK -> picks up next task
  \- Context high (65%+) -> creates handoff -> provides resume prompt -> stops

The handoff files are saved to HANDOFF_DOC/ in your project root (auto-added to .gitignore). Sensitive credentials go in .credentials/ (also auto-gitignored). When starting a new session, paste the provided prompt and the agent picks up exactly where it left off.

Session Guardian is auto-loaded for Captain and Strategist -- no configuration needed.

Installation

Prerequisites

• OpenCode installed
• Node.js installed (includes npm)

Step 1: Add to plugin config

Agent-guided install

Paste this into your LLM agent session. The agent will ask the setup questions in chat, then run the installer with --no-tui so the terminal command itself does not open a menu:

Install and configure oh-my-crew by following the instructions here:
https://raw.githubusercontent.com/michaelxer/oh-my-crew/refs/heads/oh-my-crew/docs/guide/installation.md

This URL is instructions for the agent, not the installer itself. LLM agents should fetch/read the guide with curl.exe on Windows PowerShell or curl in POSIX shells:

curl -fsSL https://raw.githubusercontent.com/michaelxer/oh-my-crew/refs/heads/oh-my-crew/docs/guide/installation.md

If you are installing directly as a human, skip the curl step and run the terminal wizard below.

If you paste the guide into OpenCode and the agent says the opencode command is missing, that usually only means the CLI is not on the agent shell's PATH. Since you are already inside OpenCode, the agent should continue the OMC setup and ask you to restart OpenCode after configuration.

Terminal wizard

If you are installing yourself, run this in a real terminal such as PowerShell, Windows Terminal, Terminal, iTerm, or your Linux shell. Do not run the interactive wizard inside an OpenCode agent/chat message, because agents usually cannot display terminal menus. The terminal wizard will ask the setup questions there:

npx oh-my-crew@latest install

or:

bunx oh-my-crew@latest install

If you want an OpenCode agent to install OMC for you, use the agent-guided prompt above instead. The agent will ask questions in chat and then run the non-interactive installer with your answers.

The npm package runs the JS installer directly, so normal npx/bunx installs do not require any separate Windows, Linux, or macOS binary package.

For scripted setup and full LLM-agent instructions, see docs/guide/installation.md. In this project, --no-tui means "no terminal menu"; it does not mean "no questions". The agent can still ask questions in chat and pass your answers as flags.

The installer supports Claude, OpenAI/ChatGPT, Gemini, GitHub Copilot, OpenCode Zen, Z.ai, Kimi, OpenCode Go, Vercel AI Gateway, and custom OpenAI-compatible providers. It asks only the setup questions needed for the provider path you choose.

At the end, the installer writes the OpenCode plugin config, writes oh-my-crew.json, and shows the generated model assignments for the crew plus the verification commands. Agent-guided installs should also summarize each crew role and the main OMC features: ultrawork/ulw, Architect/Planner mode, background agents, Session Guardian handoffs, built-in MCPs, LSP/AST-Grep tools, and model fallbacks.

Manual config is also supported.

Edit ~/.config/opencode/opencode.json (or opencode.jsonc) and add "oh-my-crew" to the plugin array:

{
  "plugin": ["oh-my-crew"]
}

Recommended: Use "oh-my-crew" only. It includes everything you need from OMO and avoids dropdown duplication plus sub-agent routing conflicts.

Important: If you keep "oh-my-openagent@latest" alongside OMC, you may get duplicate dropdown entries and old OMO sub-agent prompts can still win at runtime. For the cleanest setup, remove OMO and run OMC independently.

Step 2: Restart OpenCode

OpenCode will auto-install the plugin from npm on startup. If the crew agents don't appear, install manually:

cd ~/.config/opencode
npm install oh-my-crew --save

Then restart OpenCode again.

Step 3: Verify

opencode
opencode agent list
# You should see Captain, Strategist, Foreman, Sage, etc. in the agent dropdown/list
# Type "ultrawork" to activate all agents

Configure your proxy provider

{
  "provider": {
    "your-proxy": {
      "type": "openai",
      "url": "http://your-proxy:port/v1",
      "key": "your-api-key"
    }
  }
}

Anonymous telemetry is enabled by default to track active installations (DAU/WAU/MAU). A single event is sent at most once per UTC day per machine using a hashed installation identifier, never the raw hostname, and PostHog person profiles are not created. Disable with npx oh-my-crew install --disable-telemetry, OMO_SEND_ANONYMOUS_TELEMETRY=0, or OMO_DISABLE_POSTHOG=1. See Privacy Policy and Terms of Service.

Install from source (for developers)

Only needed if you want to modify the code:

git clone https://github.com/michaelxer/oh-my-crew.git
cd oh-my-crew
bun install
bun run build
bun link

Then add "oh-my-crew" to your opencode.json plugin array.

Upstream Features

This fork inherits all features from oh-my-openagent:

| Feature | Description | |---------|-------------| | Discipline Agents | Captain orchestrates Strategist, Sage, Scribe, Scout in parallel | | ultrawork / ulw | One word activates every agent. Runs until done | | IntentGate | Analyzes true user intent before acting | | Hash-Anchored Edits | LINE#ID content hash validates every change. Zero stale-line errors | | LSP + AST-Grep | IDE-precision refactoring for agents | | Background Agents | 5+ specialists running in parallel | | Built-in MCPs | Exa (web search), Context7 (docs), Grep.app (GitHub search) | | Ralph Loop | Self-referential loop until 100% done | | Architect Planner | Interview-mode strategic planning before execution | | Claude Code Compatible | All hooks, commands, skills, MCPs work unchanged | | Session Guardian | Auto git checkpoints, context monitoring, structured handoff docs for seamless multi-session work |

For full documentation, see the upstream project.

Syncing with Upstream

This is a proper GitHub fork. To pull latest upstream changes:

git remote add upstream https://github.com/code-yeongyu/oh-my-openagent.git
git fetch upstream
git merge upstream/dev

Credits

This project is based on:

• oh-my-opencode / oh-my-openagent by @code-yeongyu (YeonGyu Kim) -- the original architecture, agents, tools, hooks, and all core features
• oh-my-china by @enowdev -- the China-compatible fork with content filter fixes that this project builds upon

This fork renames all agents from cultural/political themed names to clean, role-descriptive crew names while preserving all upstream functionality.

• Original repo: github.com/code-yeongyu/oh-my-openagent
• China fork: github.com/enowdev/oh-my-china
• npm: oh-my-crew
• npm (original): oh-my-opencode
• License: SUL-1.0 (inherited from upstream)
• Discord: Join the community

Support

If you find Oh My Crew useful, consider:

• Give it a star on GitHub to help others discover it
• Report issues or suggest features in GitHub Issues
• Join the Discord community for discussion

License

This fork follows the same SUL-1.0 license as the upstream project. See NOTICE.md for modification details as required by the license.