| name | agentteams |
| description | Mode B parallelism. Manually invoke Claude Code Agent Teams (multiple full Claude Code instances coordinating via shared task list + worktree isolation + Agent View). The multispec engine auto-selects this mode when the work demands it; this skill is for forcing it. |
/agentteams — Mode B parallelism (Claude Code Agent Teams)
Multiple full Claude Code instances coordinating via a shared task list with worktree isolation. Different from Mode A (SDK subagents in one query()) — these are separate processes, separate sessions, separate worktrees.
Requirements
- Claude Code v2.1.32+ (
/agent-teams shipped experimental).
CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=true (set automatically by .claude/settings.json).- Agent View available (Claude Code v2.1.139+) for live monitoring.
How auto-selection works
The multispec engine picks Mode B when ANY of:
-
5 sub-Specs
- Estimated total time > 30 minutes
- Cross-sub-Spec peer coordination required
- Overlapping write sets (worktree isolation solves this)
Override with cmax run "..." --mode teams or cmax run "..." --mode solo.
Mechanics
- Each sub-Spec spawns a background Claude Code session with the SPEC + shared task list seed.
- Shared task list:
.claudemax/state/agent-teams-<ts>/shared-task-list.md. Every teammate has read/write.
- Worktree isolation: each session writes to
.claude/worktrees/<session-id>/ automatically on first write.
- Agent View dashboard:
cd && claude --agent-view shows live state of all running sessions.
When to manually invoke
- A multi-day swarm-style project where you want the coordination primitive even for < 5 sub-Specs.
- Stress-testing the agent-teams flow on a small problem before trusting it on a big one.
Anthropic's proof point
16 agents wrote a Rust-based C compiler from scratch (100K LOC, ~2000 sessions, $20K credit) using this primitive. See https://www.anthropic.com/engineering/building-c-compiler.
