Rexa DocsAgents
Skills
The skills Rexa gives agents, and turning an agent's skills, plugins and MCP servers on and off.
Rexa skills
Rexa links two skills into every supported agent. They only run inside a Rexa terminal.
| Skill | What it teaches the agent |
|---|---|
| browser | Drive a Rexa browser tab with rexa web: open, snapshot, act, wait, check. |
| handoff | Pass work to another live agent with rexa agent send, as a reviewer and worker loop of up to three rounds. |
Each agent names them its own way:
| Agent | Browser skill | Handoff skill |
|---|---|---|
| Claude Code | /rexa:browser |
/rexa:handoff |
| Codex | $rexa:browser |
$rexa:handoff |
| Cursor Agent | /rexa-browser |
/rexa-handoff |
| OpenCode | rexa-browser |
rexa-handoff |
| Pi | /skill:rexa-browser |
/skill:rexa-handoff |
Agents also pick a skill on their own when the task matches it, such as “check the page in the browser”.
Session context
When an agent starts in a Rexa terminal, a hook gives it a short note: it runs in Rexa, and can use rexa web, rexa sessions, rexa lsp, rexa pty ls and rexa extension ls. Outside Rexa the hook adds nothing.
Manage skills, plugins and MCP servers
Open the Agents panel (⌘ ⇧ A) and click an agent. Its manage tab lists what the agent loads, in tabs: MCPs, Memories, Skills, Commands, Hooks and Config. Filter by All or Enabled, or search.
For Claude Code and Codex, the tab has switches:
| Agent | You can toggle | Written to |
|---|---|---|
| Claude Code | Skills, commands, plugins | ~/.claude/settings.json, or the worktree’s .claude/settings.json or .claude/settings.local.json |
| Codex | Skills, plugins, MCP servers | ~/.codex/config.toml |
Each switch shows which file decides it, with Unset, On or Off per file. Entries set by a file Rexa does not write, such as a managed policy, are marked and read-only. Cursor Agent, OpenCode and Pi show their lists read-only.
The tab rereads when the agent’s config files change.