multi-gitter-agent 🤖

A high-performance CLI tool that combines multi-gitter with AI agents (Gemini, Claude Code, Copilot) to perform natural language refactors across one or many repositories.

Overview

multi-gitter-agent allows you to orchestrate complex code modifications across your entire organization using AI. It handles the cloning, branching, and Pull Request creation while delegating the logic to your favorite AI agent.

Installation

Shell Installer (Linux & macOS)

Install the latest binary automatically:

curl -sSL https://raw.githubusercontent.com/juftin/multi-gitter-agent/main/install.sh | bash

Via Go

go install github.com/juftin/multi-gitter-agent@latest

CLI Flags

• -p, --prompt: The natural language instruction for the agent.
• --prompt-file: Path to a file containing the instruction.
• --prompt-template: Path to a custom Go template file for the full agent prompt. Supports .UserPrompt, .Repository, .BaseBranch, and .DryRun.
• -a, --agent: (Required) The AI agent to use (gemini, claude, copilot). Can also be set via MULTI_GITTER_AGENT environment variable.
• -y, --yolo: Automatically approve all AI actions.
• -i, --interactive: Take manual decisions before committing changes.
• -C, --concurrent: Number of repositories to process simultaneously (Interactive mode requires -C 1).
• --org, --repo, --user, --topic: Standard multi-gitter repository discovery flags.
• --repo-search, --code-search: GitHub search discovery flags.
• --repo-include, --repo-exclude: Regex-based repository filtering.
• --draft: Create Pull Requests as drafts.
• -t, --token: Your platform access token (e.g., GITHUB_TOKEN).

Usage Examples

Bulk Dependency Update (Automated)

Update a specific dependency across all repositories in an organization using Gemini:

multi-gitter-agent run --agent gemini \
  --prompt "Update github.com/stretchr/testify to v1.10.0 in go.mod" \
  --org my-org --yolo

Targeted Refactor via Code Search (Interactive)

Use GitHub Code Search to find repositories containing a specific pattern and refactor them with Claude:

multi-gitter-agent run --agent claude \
  --code-search "filename:package.json lodash" \
  --prompt "Replace lodash with native Array methods where possible" \
  --interactive

Advanced Repository Discovery

Find repositories by GitHub Topics and exclude legacy projects using Regex:

multi-gitter-agent run --agent gemini \
  --topic production --topic microservice \
  --repo-exclude "^legacy-.*" \
  --prompt "Add a HEALTHCHECK instruction to the Dockerfile" \
  --yolo

Targeted Repository Search

Use GitHub's repository search syntax to target specific languages or metadata:

multi-gitter-agent run --agent gemini \
  --repo-search "topic:react language:typescript" \
  --prompt "Migrate from TSLint to ESLint" \
  --draft --yolo

Organization-wide Security Audit

Run a security audit across all repositories and only create PRs for actionable fixes:

multi-gitter-agent run --agent gemini \
  --prompt "Check for hardcoded API keys and move them to environment variables" \
  --org my-company --concurrent 5 --yolo

Complex Refactor with Prompt File

Provide detailed, multi-step instructions from a Markdown file:

multi-gitter-agent run --agent copilot \
  --prompt-file ./docs/refactor-plan.md \
  --user my-github-username --branch feature/api-v2-migration

Custom Full Prompt Template

Use your own Go template for the full worker prompt while still injecting the user request:

multi-gitter-agent run --agent claude \
  --prompt "Migrate to structured logging" \
  --prompt-template ./templates/worker-prompt.tmpl \
  --org my-org --yolo

In your template file, include {{ .UserPrompt }} where you want the user prompt injected.

Features

• Integrated Orchestration: Natively wraps multi-gitter for seamless multi-repo management.
• Agent Support: Built-in support for Gemini CLI, Claude Code, and GitHub Copilot.
• Interactive TUI: A polished Bubble Tea interface for managing per-repository tasks.
• YOLO Mode: Fully automated execution for bulk refactoring.
• Robust TTY Handling: Custom PTY wrapper ensures stability for complex AI CLIs.
• Context-Rich Prompts: Automatically provides agents with repository metadata and safety mandates.

How it works

When you execute multi-gitter-agent run, the following orchestration occurs:

1. Repository Discovery: The tool uses the provided flags (--org, --user, --repo, or --code-search) to identify all target repositories.
2. Cloning: multi-gitter-agent (via the embedded multi-gitter library) creates temporary clones of each repository and checks out a new feature branch.
3. Agent Invocation: For each repository, the tool launches an Internal Agent Runner. This runner:
   • Sets up a stable terminal environment using a Pseudo-Terminal (PTY).
   • Renders a rich Markdown prompt containing your instructions and the repository's context.
   • Executes your chosen AI agent (Gemini, Claude, or Copilot) within that PTY.
4. AI Dialogue: The AI agent analyzes the code and applies modifications. In --interactive mode, you can review and respond to the agent's thoughts. In --yolo mode, all agent actions are automatically approved.
5. Change Detection: After the agent finishes, the tool checks for file modifications.
6. PR Creation: If changes were made, the tool automatically commits the work, pushes the branch, and opens a Pull Request on your target platform (GitHub, GitLab, etc.).

Safety

• Dry Run: Use --dry-run to see what changes the AI would make without pushing anything.
• Security Mandates: The tool explicitly instructs agents to never hardcode secrets or modify files outside the requested scope.