Agent Adapter

Watch your local AI coding agents (Claude Code · Codex · Cursor · Gemini CLI · OpenClaw · Hermes) in real time — working / idle / waiting — and react back (answer an approval, confirm, prompt, interrupt).

It runs headless on your machine and uplinks to a Commander, where you see status and send commands back.

---

How it works

 agents → hooks ──(local socket)──▶ ADAPTER ──(WSS)──▶ Commander
   ▲ status (working/idle/waiting)        │              (your dashboard)
   ▼ react back (answer/confirm/…)        └─ state + react-back

1. Listen — a tiny hook in each agent posts an event to a local socket on every action.
2. Track — events fold into one live status per session: working, idle, or waiting (an approval or question).
3. React back — a command for a waiting session is typed into the agent (answer / confirm / prompt / interrupt).
4. Uplink — status streams up to your Commander, and commands stream back down.

---

Install

Needs Node ≥ 22 and git.

# macOS / Linux
curl -fsSL https://nohope88.github.io/agent-adapter/install.sh | bash

# Windows (PowerShell)
irm https://nohope88.github.io/agent-adapter/install.ps1 | iex

The script clones, builds, detects your installed agents, wires their hooks, and registers a background daemon (launchd / systemd / Scheduled Task). Re-run it any time to update.

Uninstall: ./install.sh --uninstall (.\install.ps1 --uninstall on Windows).

---

Use

You only need these commands:

aca login --token <cmdr_ak_…>   # connect to your Commander — the daemon starts uplinking
aca logout                      # disconnect (remove the stored credential)
aca verify                      # re-scan agents after you install or remove one
aca setup                       # re-detect agents and wire hooks (the installer runs this for you)

After aca login, everything is automatic: the daemon reports status up and applies commands coming down. Watching status and reacting back happens in your Commander/dashboard, not the local CLI.