Gemini-GPT Hybrid workflow
Two Claude Code subagents: broad tasks go to Gemini, narrow ones to GPT, and results return to Claude for review.
- 153 ★ on the repository
- Last commit
- MIT
Broad access
Why this level
- The `hard` mode lets external models write to project files with no check before the write.
- Project code is sent to Gemini and GPT.
Install
- Works in
Manual install
git clone https://github.com/NEWBIE0413/gemini-gpt-hybrid.git
cd gemini-gpt-hybrid
cp -r agents ~/.claude/Commands from the Setup section of the README for macOS and Linux. Windows uses `xcopy`.
Access and checks
How we reviewBroad accessThe tool can change data, run commands or handle secrets.
- The `hard` mode lets external models write to project files with no check before the write.
- Project code is sent to Gemini and GPT.
This is third-party code. Review the repository files before installing.
What it does
The repository holds two subagent files for Claude Code. The agent reads the scope of a task and picks a model. Whole-repo audits and architecture reviews go to Gemini with its 1M+ token context. Single-file debugging and focused features go to GPT through Cursor Agent. The `soft` mode keeps code writing with Claude: Gemini and GPT only analyze, and Claude reviews their output for vulnerabilities and quality. The `hard` mode lets external models change files directly, with no check before the write.
Who it is for. Claude Code users with Gemini CLI and Cursor Agent access who want one entry point for tasks of different size.
Good fit when
- A security audit across the whole repository does not fit one model's window.
- A single function fails and you want a fast loop without full context.
- Only Claude writes production code while external models provide analysis.
- An experiment on a branch where rough edits and a Git rollback are acceptable.
Not a fit when
- Gemini CLI or Cursor Agent are not installed and configured.
- Code must not be sent to third-party models.
- The task is small and Claude Code alone is enough.
Example request
@gemini-gpt-hybrid perform a security audit of the whole repository and give a list of fixes.Limitations
Claude Code, Gemini CLI (`npm install -g @google/gemini-cli`) and Cursor Agent for GPT are required; on Windows the README asks for WSL. Project code goes to Gemini and GPT, which matters for private repositories. The `hard` mode can overwrite work: commit and branch first. The `soft` review is a layer, not a guarantee, so read the diffs yourself. The last commit in the repository is from July 2026.
How to disable. Delete the two `gemini-gpt-hybrid-*.md` files from `~/.claude/agents/`. Gemini CLI and Cursor Agent stay and are removed separately.
FAQ
How does soft differ from hard?
In `soft` only Claude writes code after review, in `hard` external models change files directly. Use `soft` for production.
How do I call the subagent?
Mention it in a conversation: `@gemini-gpt-hybrid` and the task. It picks Gemini or GPT by request size and calls both in sequence for mixed tasks.
What must be installed first?
Claude Code, Gemini CLI and Cursor Agent for GPT access. The README advises WSL on Windows.
How do I roll back hard mode edits?
Only through Git. Commit and create a branch before running, as the `hard` section of the README says.
Which file should I choose?
`gemini-gpt-hybrid-soft.md` for working code, `gemini-gpt-hybrid-hard.md` for prototypes. More in `AGENT_SELECTION_GUIDE.md`.
Discussion
No questions or reviews yet.
Related
WorkflowsClaude Code skills that delegate UI to Gemini via agy and code to Codex, with cross-review and parallel git worktrees.
A Claude Code plugin that hands bulk work to Gemini through the Antigravity CLI and leaves decisions and verification to Claude
Claude Code calls Gemini to analyse whole repositories, search with Google Search grounding and read screenshots.
A personal ~/.claude for Claude Code: twelve subagents, review commands, worktree and goal skills and a Go status line.
System instructions and targeted reminders for GPT in Codex: fewer ceremonies, more decisions and careful craft
Plugin and MCP server that asks GPT, Gemini, Grok and local models to review a plan or hunt for bugs before code ships.
README in short
Routing follows the scope of the request, and the decision to write files depends on the chosen mode.