| name | team-run |
| description | Orchestrate multiple specialized subagents in a coordinated workflow. Use when running multiple agents together, delegating to agents, or executing complex multi-agent pipelines. |
Team Run
Trigger: Use /team or when the user mentions running multiple agents together,
or asks to "use the team", "run agents", "delegate to agents".
Purpose
Orchestrates multiple specialized subagents in a coordinated workflow. Each agent
has a specific role, model assignment, and tool access defined in .qwen/agents/.
Agent Directory
| Agent | File | Role | Tools |
|---|
| sisyphus | sisyphus.md | Main ultraworker — decomposes & executes complex tasks | All tools |
| prometheus | prometheus.md | Strategic planner — creates phased execution plans | Read + write files, search |
| hephaestus | hephaestus.md | Deep implementation — builds features end-to-end | All tools |
| atlas | atlas.md | Architecture analyst — reviews structure & design | Read-only + search |
| explore | explore.md | Researcher — finds code, maps architecture | Read-only + search + web |
| librarian | librarian.md | Documentation specialist — writes docs, READMEs | Read + write + edit |
| code-reviewer | code-reviewer.md | Quality gate — reviews for bugs, security, perf | Read-only + search |
| testing-expert | testing-expert.md | Test specialist — writes & fixes tests | All tools |
Orchestration Patterns
Pattern 1: Pipeline (Sequential)
explore → prometheus → hephaestus → testing-expert → code-reviewer → librarian
Use for: Complete feature from scratch with full quality gates.
Pattern 2: Fan-out (Parallel)
→ hephaestus (feature A)
root → hephaestus (feature B) → code-reviewer
→ librarian (docs)
Use for: Independent features that can be built in parallel.
Pattern 3: Review Loop
implement → code-reviewer → fix issues → re-review
Use for: Ensuring high quality on critical code.
Pattern 4: Research → Plan → Execute
explore → atlas → prometheus → sisyphus → testing-expert
Use for: Large projects on unfamiliar codebases.
How to Run
- Identify the pattern needed for the task
- Invoke subagents by name in your prompt or commands:
- "Use the sisyphus subagent to..."
- "Let the testing-expert write tests for..."
- Wait for each subagent to complete before aggregating
- Report unified results
Model Assignment Reference
Agents use the model configured in settings.json modelProviders. The current
setup assigns all agents to the qwen-oauth provider with model selection handled
by Qwen Code's model resolution.
Delegation Hints
Include these phrases in your commands to encourage proper delegation:
- "Delegate to [agent-name]"
- "Use the [agent-name] subagent"
- "Let [agent-name] handle this"
- "Run [agent-name] to..."
