@zktable/agents
v0.1.1
Published
zkTable AI agent harness: game-agnostic Agent interface, HeuristicAgent/RandomAgent baselines, and ClaudeAgent. Agents see only a PlayerView (public state + their own secret) and move through the same proof path as humans.
Readme
@zktable/agents
The zkTable AI-agent harness: a game-agnostic Agent interface, the
HeuristicAgent / RandomAgent baselines, and ClaudeAgent. Every agent sees
only a PlayerView (public state + its own secret) and moves through the exact
same ZK proof boundary as a human — so an AI opponent is mathematically
incapable of cheating.
Part of zkTable — a TypeScript SDK for building trustless, privacy-preserving board games on Stellar.
Install
npm install @zktable/agents @zktable/coreWhat's inside
Agent— the interface every policy implements: given aPlayerView, return a legal move.HeuristicAgent/RandomAgent— deterministic baselines that need no API key; the default for local play and tests.ClaudeAgent— an LLM-backed policy. SetANTHROPIC_API_KEYin the environment (server-side only — the key never reaches the browser).
Because agents only ever receive a PlayerView, they cannot see another
player's secret, and their moves are ZK-verified on-chain like any other.
License
MIT — see LICENSE.
