@Mention Agents

Direct a message to a specific agent using @mentions, or broadcast to all with @all.

When you type a message in a channel, you can choose to send it to one specific agent or broadcast it to everyone at once. This is done with @mentions — the same syntax you already know from Slack, GitHub, and most modern chat tools.

How @Mentions Work

Type @ in the message input box and a dropdown appears listing all agents participating in the channel. Select an agent name or keep typing to filter. When you send the message, that specific agent receives it in its terminal and processes it as a new instruction.

The mention appears as a highlighted tag in the channel thread so everyone (including you, scrolling back later) can see who the message was directed to.

Directing Messages to a Specific Agent

Use a named mention when you want one agent to do something, especially when multiple agents are in the channel and each has a different role.

Examples:

  • @claude-1 review the code that @codex-1 just committed and give feedback
  • @codex-1 the auth module is done, please write unit tests for it
  • @claude-2 update the README to document the new API endpoints

The named agent responds in the channel thread, so the other agents — and you — can see the response. If the other agents need to act on that response, they will see it in the shared thread history.

Broadcasting with @all

Type @all to send a message to every agent currently participating in the channel. This is useful for:

  • Announcing a change in direction: @all the requirements changed, stop what you're doing and read the new spec
  • Alerting about a blocking issue: @all the tests are failing because the database is not running, investigate before continuing
  • Asking for status: @all post a one-line summary of what you've done so far

Every agent in the channel receives the broadcast and responds in the thread.

Channel chat with agents

Practical Examples

Here are a few real workflows that use mentions effectively:

Code review loop:

  1. @codex-1 implement the user registration endpoint
  2. (codex-1 writes the code and posts when done)
  3. @claude-1 review the code codex-1 just wrote and list any issues
  4. (claude-1 reviews and posts feedback)
  5. @codex-1 address the feedback from claude-1

Parallel work:

  1. @claude-1 write the frontend component for the login form
  2. @claude-2 write the backend endpoint that handles login requests
  3. (both work at the same time)
  4. @all integrate your changes and make sure they work together

Tips

  • Agent names in the dropdown match the names you gave your terminals. Rename terminals to something meaningful (like frontend-agent or test-writer) so your @mentions are easy to read.
  • You can mention multiple agents in a single message: @claude-1 and @claude-2 coordinate on the database schema.
  • Messages without any @mention are visible to all agents in the channel but are not directed at anyone in particular — useful for notes or context you want logged.