@neurosquad/core

The integration core shared by nsq and the NeuroSquad desktop app: how to start Claude Code, Codex and OpenCode with a per-agent configuration layer, how to learn their exact status from their own hooks, and how to read what they spent.

Node.js ≥ 22.13, no Electron, no DOM. node-pty is an optional peer dependency (only PtyHost needs it).

| Area | Modules | | --------- | ---------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Launch | prepareLaunch → { command, args, env, … } per harness; the user's own CLI config is never written | | Status | machine (pure state machine), hub (hook payloads, timers, the Claude transcript), hooks/server (loopback endpoint with per-agent tokens) | | PTY | PtyHost (spawn with ConPTY on Windows, trust prompt, session-not-found recovery, submit/paste/interrupt), events, screenMirror | | Providers | OpenRouter recipes and attribution headers; the user's own OpenAI/Anthropic-compatible servers (validation, connection test, model lists, per-harness recipes, a Responses ⇄ chat completions gateway for Codex) | | Usage | Claude Code / Codex / OpenCode log readers, exact pricing (agentCost) | | Git | per-agent worktrees |

See docs/harnesses.md and docs/status.md.

MIT licensed.