frankcode 0.1.0

Advanced multi-agent orchestrator for OpenCode

FrankCode

Advanced multi-agent orchestrator for OpenCode.

FrankCode acts as a wrapper and orchestration layer around OpenCode, managing specialized agent setups, safety plugins, verification passes, and environment configurations to accomplish complex development tasks safely and robustly.

Requirements

Before installing FrankCode, ensure you have the following prerequisites installed on your system:

• Python: >= 3.10
• Node.js & npm: Required to install OpenCode plugins and manage the environment dynamically. FrankCode will attempt to automatically resolve Node via nvm if available.
• OpenCode: Make sure the opencode CLI is installed and available in your PATH.

Installation

You can install FrankCode directly via pip:

pip install frankcode

Alternatively, to run from source:

git clone https://github.com/joeyism/frankcode.git
cd frankcode
pip install -e .

Usage

FrankCode provides a simple CLI. To execute a task using the default multi-agent orchestrator:

frankcode "Create a new React component that fetches data from an API and displays it in a table"

Options

• --model: Specify the LLM model to use (default: google/gemini-3.1-pro-preview).
• --verbose, -v: Show debug logs.
• --no-verify: Disable the autonomous verification and fix passes.
• --clean: Wipe the .frankcode execution environment before starting.

Example:

frankcode "Refactor the auth logic to use JWT" --model openai/gpt-5.4 --verbose

How It Works

1. Environment Setup: FrankCode creates an isolated .frankcode directory to store logs, quarantine artifacts, and install specific OpenCode plugins dynamically.
2. Execution: Your prompt is passed into an enhanced OpenCode session. FrankCode actively monitors the process for infinite error loops or idle timeouts.
3. Verification Loop (Optional but enabled by default): After the main task completes, an adversarial validation agent reads the outputs and devises independent tests. If the tests fail, FrankCode triggers a "fix pass" and repeats the verification.
4. Cleanup: Temporary artifacts generated during agent reasoning are quarantined.

License

MIT License. See LICENSE for details.