@philtzjp/omni-subagents
v0.1.2
Published
Claude Code subagents for multimodal capabilities via OpenCode + OpenRouter
Maintainers
Readme
@philtzjp/omni-subagents
Claude Code subagents that call external AI models via OpenCode + OpenRouter. Fills Claude Code's multimodal gaps by delegating to specialized models.
Subagents
| Name | Model | What it does |
|---|---|---|
| omni-vision | Gemini 3.1 Pro | Image/screenshot analysis — fine-grained UI inspection, OCR, diagram interpretation |
| omni-codex | GPT-5.3 Codex | Code second opinion — alternative implementations, blind spot detection |
| omni-imagegen | GPT-5 Image | Image generation — diagrams, UI mockups, charts, illustrations |
| omni-researcher | Gemini 3.1 Pro (2M) | Large-context research — cross-file analysis, massive document processing |
How it works
Each subagent is a lightweight relay (runs on Haiku) that executes opencode run -m <model> under the hood. Claude Code invokes them automatically via the Agent tool when it determines an external model would handle the task better.
User → Claude Code → Agent(subagent_type: "omni-vision") → opencode run → OpenRouter → Gemini 3.1 ProInstall
npx @philtzjp/omni-subagents installThe installer will:
- Check that
opencodeCLI is installed - Prompt for your OpenRouter API key (if not already configured)
- Copy subagent definitions to
~/.claude/agents/
Commands
npx @philtzjp/omni-subagents install # Install subagents + configure key
npx @philtzjp/omni-subagents setup # Set or update OpenRouter API key
npx @philtzjp/omni-subagents uninstall # Remove subagents
npx @philtzjp/omni-subagents list # Show installed subagentsRequirements
- Claude Code
- OpenCode CLI with OpenRouter credentials
- OpenRouter API key with access to the target models
License
MIT
