Claude Code Agent Teams + Skills: How to Build Multi-Agent Workflows (2026)

Quick Answer: Claude Code Agent Teams let multiple Claude sessions work together on the same project. Each teammate loads skills from your .claude/skills/ directory automatically. Combine Agent Teams with specialized skills to build parallel workflows: one teammate runs code review, another writes tests, a third handles documentation. Skills make each teammate a specialist instead of a generalist. Browse specialized skills at agensi.io/skills.

Claude Code Agent Teams launched in February 2026 alongside Opus 4.6. It's the most ambitious feature Anthropic has shipped for Claude Code: multiple autonomous Claude sessions that coordinate, message each other, and divide work in parallel.

The difference from subagents is communication. Subagents run within a single session and report results back to the parent. They can't talk to each other. Agent Teams removes that bottleneck. Teammates message each other directly, claim tasks from a shared list, and challenge each other's findings.

Skills make Agent Teams dramatically more effective. Without skills, every teammate is a generalist. With skills, each teammate becomes a specialist.

How skills load in Agent Teams

When a lead agent spawns a teammate, the teammate loads the same project context as a regular Claude Code session. This includes your CLAUDE.md, MCP servers, and every skill in your .claude/skills/ directory.

The teammate receives the lead's spawn prompt but not the lead's conversation history. This means the teammate starts fresh with its assigned task, but with the full skills library available.

Each teammate independently matches tasks to skills. If you spawn a teammate with "review the authentication changes," it loads your code-review skill automatically. If another teammate gets "write tests for the payment module," it loads your testing skill.

See SKILL.md in action

attack-tree-construction

by LocoLoboZ

$10

User Detection & Session Context Reviewer

by LB Creations

$14.99

ai-team-builder-kanban

by Roy Yuen

$5

Browse all skills

Best skill combinations for a 3-agent team

The most effective Agent Teams setup uses three specialized teammates. Here's a configuration that works well for shipping features.

Teammate 1: Code Reviewer. Spawned with a code-review skill that checks logic errors, security vulnerabilities, performance issues, and style violations. This teammate reviews the lead's implementation as it's written. It can message the lead directly when it finds an issue.

Teammate 2: Test Writer. Spawned with a testing skill that generates tests matching your framework and conventions. This teammate writes tests in parallel with implementation. It reads the lead's changes through the shared task list and generates corresponding tests.

Teammate 3: Documentation. Spawned with a documentation skill that generates API docs, inline comments, and changelog entries. This teammate updates documentation as the other teammates implement and test.

The lead coordinates all three, makes implementation decisions, and synthesizes their findings. The result: a feature ships with code review, tests, and documentation in the time it used to take to write just the implementation.

Practical example: shipping a feature

Here's what this looks like in practice.

You tell the lead: "Implement the credit deduction endpoint. It should validate the user's balance, deduct the credits, log the transaction, and return the updated balance."

The lead creates a task list: implement the endpoint, write tests, update API documentation, review for security issues.

The lead starts implementing. Teammate 1 (reviewer) picks up the security review task and begins analyzing the implementation as it takes shape. Teammate 2 (tester) claims the testing task and generates test cases based on the endpoint's requirements. Teammate 3 (documenter) claims the documentation task.

When the reviewer finds an issue ("the balance check and deduction aren't atomic, this has a race condition"), it messages the lead directly. The lead fixes it. The tester's tests already cover the race condition because the testing skill explicitly checks for concurrency edge cases.

Total time: roughly the same as implementing without a team. But the output includes reviewed code, comprehensive tests, and updated documentation.

Cost considerations

Agent Teams consume tokens linearly. Three teammates cost 3x the tokens of a single session. At Anthropic's reported average of $13 per developer per day for single-session use, a three-agent team could cost $30 to $40 per day.

Skills help offset this cost. A teammate with a focused skill completes its task faster because it doesn't explore multiple approaches. The testing teammate with a testing skill writes the correct test structure on the first attempt instead of iterating. Fewer tokens per task, even with more agents running.

Whether the cost is worth it depends on your workflow. For shipping features that need review, tests, and docs, Agent Teams with skills can compress a full day of work into an hour. The token cost is real but the time savings are larger.

Skills that work best with Agent Teams

Not every skill is equally useful in a team context. The best team skills are focused, independent, and produce discrete output that other teammates can consume.

Code review skills work well because the reviewer operates independently on already-written code. Testing skills work well because tests are a self-contained deliverable. Documentation skills work well for the same reason.

Skills that require deep back-and-forth conversation (architecture planning, debugging sessions) work better in the lead agent's session, not as a teammate task.

Browse skills organized by task at agensi.io/skills.