Docs
Harnesses
A harness is the coding agent a session runs: Claude Code, Codex, Gemini CLI and the rest. Slashcode installs none of them and resells nothing: it starts the ones already on your computer, signed in with your own accounts.
Built in
The daemon knows these 33 harnesses by name. Each one is offered in the app as soon as its command is installed on the computer.
| Harness | id | Command it runs |
|---|---|---|
| Claude Code | claude | claude |
| Codex | codex | codex-acp |
| Gemini CLI | gemini | gemini --acp |
| GitHub Copilot | copilot | copilot --acp |
| Cursor | cursor | agent acp |
| OpenCode | opencode | opencode acp |
| Grok | grok | grok agent stdio |
| Qwen Code | qwen | qwen --acp |
| Kimi Code | kimi | kimi acp |
| Mistral Vibe | vibe | vibe-acp |
| Kiro | kiro | kiro-cli acp |
| Auggie | auggie | auggie --acp |
| Cline | cline | cline --acp |
| Factory Droid | droid | droid exec --output-format acp |
Also built in: OpenHands, Docker Agent, Qoder CLI, Blackbox AI, Hermes Agent, OpenClaw, fast-agent, Stakpak, VT Code, Command Code, Pi, Poolside, CUGA, Imagera Code, Mercury CLI, Octomind, Raven, Raxol, ZenCODE.
See what a computer has
$ slashcode agentsOne line per harness: its id, name and command, then available or not installed, and accepts edits or accepts all when you let it act without asking.
The daemon looks for commands on the PATH of the shell that ran slashcode service install. If you install a harness somewhere new and it still shows not installed, run slashcode service install again from a shell that finds it (this restarts the daemon).
Add any ACP agent
Any program that speaks the Agent Client Protocol on its stdin and stdout can be a harness. Give it an id (lowercase letters, digits, - and _), a name for the app, and its command after --:
$ slashcode agent add my-agent --name "My Agent" -- my-agent --acpAdded my-agent: my-agent --acp$ slashcode agent remove my-agent
The running daemon picks it up at once. Adding an id that is already built in replaces that harness's command on this computer, which is how you point one at another install.
What a harness may do without asking
By default, every time an agent wants to edit a file or run a command, your devices ask you. Per harness and per computer, you can let some of it through:
$ slashcode agent permissions claude editsask: every request goes to your devices. The default.edits: edits to files of the session's folder are accepted, everything else still asks.all: every request is accepted without asking.
It applies from the next request, without a restart. In the app, /permissions in any session of that harness changes the same setting. See answering permissions.