Statewright – Visual state machines that make AI agents reliable
Protocol-level tool enforcement prevents jailbreaks better than system prompts.
Tmux plugin that gives visual feedback for AI agent states (running/needs-input/done). Supports Claude Code, Codex, and custom agents. Pane borders, window title colors, status bar icons.
If you run multiple LLM sessions in tmux, this stops the annoying habit of constantly switching panes by surfacing state through pane borders, window-title colors and status-bar icons. It hooks into Claude Code and Codex lifecycle events (or accepts calls to agent-state.sh), offers a one-command installer, a Knight Rider running animation, and a process-detection fallback — small, clever tmux trickery that actually improves day-to-day ergonomics for its niche.
Developers who run AI agents (Claude, Codex, custom agents) inside tmux panes and want visual notifications without switching panes
This plugin hooks into agent lifecycle events and changes pane borders, window titles, and status bar icons to reflect three states: running, needs-input, done. Colors reset when you focus the pane.
Works with Claude Code and Codex out of the box. Any other agent can integrate by calling a single shell script.
Protocol-level tool enforcement prevents jailbreaks better than system prompts.
Forces 13B models to solve SWE-bench tasks by making the problem smaller, not the model bigger.
Tmux session switcher for Claude instances, but the niche is Claude Code users with 5+ parallel agents.
Canvas-based state verification solves agent hallucination, but competes with established RPA platforms.
Two-commit spec repo with no working implementation or demo.
Sprite personality makes agents feel less abstract than logs.