Claude Code: Named Subagents Now Appear in @ Mention Typeahead
Claude CodeClaude Code v2.1.88 adds named subagents to the @ mention typeahead suggestions, making it significantly easier to invoke custom agents directly from the chat interface. Users who have defined custom subagents in .claude/agents/ or ~/.claude/agents/ can now discover and mention them in the same way they @-mention files, reducing friction in multi-agent workflows. This complements the existing subagent system — where Claude Code can spawn autonomous child processes with their own context windows, tools, and instructions — by surfacing those agents inline as first-class citizens of the prompt interface.
Key Takeaways
- Named subagents now appear in the
@mention typeahead, allowing users to invoke custom agents by name the same way they reference files — reducing friction for multi-agent workflows. - Custom agents are defined as Markdown files with YAML frontmatter in
.claude/agents/(project) or~/.claude/agents/(user), with configurable tools, model, and system prompt. - The change makes subagent delegation intentional rather than implicit — users can now explicitly route tasks to a specialized agent rather than relying on Claude Code to select one automatically.
- Subagents are distinct from Agent Teams: subagents report directly back to the orchestrator and do not talk to each other, making them lighter-weight and lower-cost for focused delegation tasks.
- The
@mention pattern is a deliberate UI design — it mirrors file mentions, making subagents discoverable in the flow of normal prompting without requiring separate commands or menus. - Combined with the v2.1.88 reliability fixes, this release materially advances Claude Code's multi-agent ergonomics: more stable long sessions plus more accessible agent invocation in the same update.
Sources & Mentions
4 external resources covering this update
Named Subagents in @ Mention Typeahead
Claude Code has supported custom subagents for some time — user-defined agents stored as Markdown files with YAML frontmatter in .claude/agents/ (project-scoped) or ~/.claude/agents/ (user-scoped). These agents can be configured with their own tool permissions, model selection, effort level, and system prompt, enabling users to build specialized helpers like a code-reviewer, test-runner, or docs-writer.
In v2.1.88, Anthropic adds those named subagents to the @ mention typeahead suggestions. When you type @ in the Claude Code chat input, your defined custom agents now appear alongside files and directories in the autocomplete list — making it possible to invoke a specific agent by name without needing to remember the exact frontmatter syntax or navigate menus.
What This Changes in Practice
Previously, invoking a custom subagent required Claude Code to discover and select it based on the task description, or the user had to explicitly describe which agent to use in natural language. With named subagents surfaced in the typeahead, the interaction becomes more intentional:
- Type
@code-reviewerto directly invoke your code review agent on the current diff - Type
@test-runnerto hand off a failing test investigation to a specialized subagent - Type
@docs-writerto delegate documentation tasks without context switching
This mirrors how @ file mentions work — a deliberate, discoverable, low-friction way to reference something by name.
Subagent Architecture Context
Claude Code subagents are distinct from Agent Teams. Subagents speak directly with the orchestrating agent, operate in their own context window, and report results back — they do not communicate with each other. A subagent is defined with a Markdown file like:
With the typeahead addition in v2.1.88, these agents become first-class citizens of the Claude Code interface rather than background infrastructure the model selects invisibly.