agent-teamlisted

# Agent Team — Hierarchical Engineering Team Simulation (HETS) Reusable toolkit primitive for orchestrating multiple agents in a tree structure that mimics a real software engineering team: PM → Senior → Mid → Junior. Each level has scoped responsibilities, contracts that must be satisfied, and budget allocations. Trust builds incrementally so high-trust roles get spot-checked rather than fully reviewed. ## When to use HETS - Multi-step tasks that decompose naturally into parallel sub-tasks - Tasks where verification matters (not just output, but approach quality) - Tasks where you'd benefit from review at multiple abstraction levels - Audits, code reviews, large refactors, multi-file features ## When NOT to use HETS - Single-actor tasks (just spawn one Agent directly) - Tasks under 30 minutes (the orchestration overhead exceeds the work) - Tasks where a single perspective is correct (no review needed) ## The role hierarchy | Real-world role | Agent role | Depth | Verifies | Budget | |----------------|-----------|-------|----------|--------| | Product Manager | super-agent | 0 | Cross-area patterns, business outcome | Uses parent context | | Senior Engineer | orchestrator | 1 | Architectural soundness of children | 8K tokens | | Mid Engineer | sub-orchestrator | 2 | Implementation quality of children | 8K tokens | | Junior Engineer | actor | 2-3 | Functional outputs match contract | 30K tokens | **Recursion limit**: `max_depth = 3` by default. At depth 3, only actors