Claude Code and Cursor through hooks installed in one click, and Codex by reading the session files it already writes, with no configuration at all. Adding an agent never means giving up a permission.
Updated 24 September 2026
| Agent | How | Setup | Plan gauge |
|---|---|---|---|
| Claude Code | Its own hooks, over a local socket | One click | Yes |
| Cursor | Its own hooks, over a local socket | One click | Not yet |
| Codex | Reads the session files it already writes | None | Yes |
The hook installers write to the agent's own configuration file and keep a backup of what was there before. Nothing is patched, injected, or wrapped.
The plan gauge in the menu bar menu shows what is left of each window of your plan and when it resets. It is asked of Claude Code and Codex themselves, under your own sign-in. Cursor does not report a quota yet.
Cursor reports working and finished. It emits no signal for "waiting on you", so Shiplings cannot show that state for Cursor, and does not pretend to. A companion that invents a state it cannot actually detect is worse than one that admits the gap.
The same character also follows what happens after the agent stops: GitHub pull requests, pushes and stars, and Vercel or Netlify deploys going out, landing green, or turning red. Tokens are stored in the macOS keychain and are read only.
No. There is no camera, microphone, screen recording, accessibility or full disk access anywhere in the app.
It will not be watched yet. A new source is one file for us, so ask and it can be added.
Each one is independent, and each can be disconnected without touching the others.