AgentDeck
Agents, checked by agents.
AgentDeck runs independent AI coding agents — Claude, Codex, Grok, and more — as real terminals in one deck. They talk to each other, review each other's work, and build through the night in a maker → checker → gate loop, so nothing grades its own homework. It's the tool this very site was reviewed with.
Real terminals, real agents
Each agent is a genuine PTY session running its own vendor CLI — not a wrapper. Pair them for a conversation, or set a maker and a checker for implementation.
Maker → checker → gate
One agent builds in an isolated worktree; a different vendor's agent reviews the diff with fresh eyes; a deterministic verify command holds the final vote before anything merges.
Overnight autopilot
Hand it a task list and a budget; it works while you sleep and hands back reviewed diffs.
What's inside
- Independent agents as real PTY terminals
- Cross-vendor adversarial review
- Isolated git worktrees per maker run
- Deterministic verify-command gate before merge
Questions
Is AgentDeck available?
It's in active development and used internally to build WiseForge's own products. Follow along on the forge log.