Adversarial multi-agent planning and deterministic pipeline execution for agents

Forge brings rigorous, adversarial architectural planning and reliable step-by-step execution to AI-assisted development. It works with Claude Code, Codex CLI, Cursor, or directly via API — the provider is pluggable.

How It Works

Planning — Two independent AI architects design competing solutions. Adversarial critics tear each design apart. Refiners address the critiques. A judge synthesizes the best elements into a final plan. The result: plans that have been stress-tested before a single line of code is written.

Execution — A deterministic pipeline runs your plan through configurable steps (code, build, lint, test, review, PR) with evidence-based validation at each stage. State lives on disk, not in LLM memory — every step gets a fresh agent with zero context rot.