@agent autocomplete for agent teams teammate selection
Feature Request
Problem
When using agent teams with multiple custom agents defined in ~/.claude/agents/, there's no way to discover or autocomplete agent names when typing @. Users must memorize all agent names to invoke them, which doesn't scale as the roster grows.
Current Behavior
- Typing
@in the CLI does nothing — no suggestions, no dropdown, no tab-completion - The only way to see available agents is
/agentscommand (separate action) Shift+Up/Downonly works after teammates are already running- No discoverability for new users or large agent rosters
Proposed Solution
When typing @ in the Claude Code input, show an autocomplete dropdown/list of:
- Custom agents from
~/.claude/agents/(name + description from frontmatter) - Active teammates currently running in the session
- Built-in agent types (if applicable)
Similar to how GitHub, Slack, and Discord show @mention suggestions as you type.
Example UX
> @
┌──────────────────────────────────────────────┐
│ @orchestrator Brain — plans & delegates │
│ @kimi-coder Primary coding agent │
│ @code-reviewer Reviews code for bugs │
│ @swift-reviewer iOS/Swift code review │
│ @debugger Diagnoses errors │
│ @test-runner Runs & fixes tests │
└──────────────────────────────────────────────┘
Typing @co would filter to @code-reviewer.
Context
- Agent teams is behind
CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1 teammateMode: "tmux"enables split-pane swarm view- Custom agents are defined as markdown files with YAML frontmatter in
~/.claude/agents/ - The
nameanddescriptionfields from frontmatter provide all the data needed for autocomplete - This becomes critical as users build larger agent rosters (6+ agents)
Additional Context
Using the tmux swarm view with multiple specialized agents (orchestrator, coders, reviewers). The split-pane UI is excellent for observing agents work in parallel, but the lack of @-autocomplete makes it hard to quickly message or invoke the right teammate without memorizing names.
This issue has 2 comments on GitHub. Read the full discussion on GitHub ↗