Conclave

A CLI tool that orchestrates multiple LLM agents to systematically audit codebases for security vulnerabilities.

Warning

This is an entirely vibe-coded project. No humans have reviewed this code. It was generated through AI-to-AI conversation and should be treated with appropriate caution.

• Do not run this on sensitive systems without review
• Do not trust the security of this tool itself
• The agents run with --full-auto flags and can execute arbitrary commands
• State files in .conclave/ contain unvalidated LLM output

Use at your own risk.

What It Does

Conclave runs a multi-stage security audit pipeline:

1. Plan - Analyzes a codebase and breaks it into subsystems
2. Assess - Spins up 3 parallel agents to review a subsystem for vulnerabilities
3. Convene - Has agents debate and refine their findings
4. Complete - Synthesizes final results

Quick Start

go build ./cmd/conclave
./conclave run                              # Codex (default)
./conclave --claude run                     # Claude
./conclave --claude=opus run                # Claude with specific model
./conclave --claude=sonnet --gemini run     # Both with Claude using sonnet
./conclave --claude --codex --gemini run    # All three

Additional Flags

./conclave run --web                        # Open web dashboard for monitoring
./conclave run --gist                       # Create secret gist of final report
./conclave run --web --gist                 # Combine both features

When multiple providers are enabled, parallel agents are distributed across them. If one provider errors or hits rate limits, agents automatically fail over to another.

Model configuration is shown in output:

Providers: Claude (opus), Gemini

Requirements

• Go 1.21+
• One of: Codex CLI, Claude CLI, or Gemini CLI