Forge Protocol

What Forge Protocol does

Forge Protocol is built on the idea that default AI assistants answer fast and let your own skills fade. Instead of giving answers, it asks better questions, critiques your draft rather than rewriting it, and stress-tests your ideas. The README says the design comes from the research programme of Federico Cabitza's lab at Milano-Bicocca, and that the four modes map onto protocols validated there. The maintainer is Lorenzo Famiglini.

Forge is the thinking partner: it asks questions, never gives answers and forces you to reason, and it is switched on with /forge-mode. Anvil is the editor and critic: it rates your draft, finds weaknesses and never rewrites, using /anvil-mode. Crucible is the idea stress-tester. Executor is the mode where AI-first help is acceptable. The README also links sections for measuring cognitive sovereignty, for schools and universities, and for the architecture, and keeps a RESEARCH.md that maps each mode to its source studies.

Key features

• Forge mode: a thinking partner that asks questions and never gives answers
• Anvil mode: an editor that rates drafts and finds weaknesses without rewriting
• Crucible mode: an idea stress-tester
• Executor mode for tasks where AI-first help is acceptable
• Slash commands such as /forge-mode and /anvil-mode
• RESEARCH.md mapping each mode to published studies

When to use it

• Having an agent critique your draft without rewriting it
• Working through a problem with an agent that only asks questions
• Introducing a structured human-AI collaboration protocol in a classroom or team

Who it is for: People and educators who want an AI assistant that keeps their own reasoning sharp instead of replacing it.

How it fits with Hermes Agent

The repository is tagged hermes-agent and shows a Hermes Agent badge in its README, and it is also tagged claude, so it is not limited to Hermes.

FAQ