team-work

Team Work

Coordinate specialist software engineering work without over-dispatching. Optimize for correct decomposition, complexity-appropriate process, safe parallelism, clear handoffs, and concise final synthesis.

Core Workflow

1. Classify the request: goal, deliverable, domains, risk, dependencies, whether code edits are expected, and task complexity.
2. Match process depth to complexity. Simple tasks stay direct; complex tasks get phased coordination. If a process feels larger than the request, downshift complexity before adding agents.
3. Read routing-matrix.md when role selection is unclear.
4. Build dependency batches before dispatching work. Run independent batches in parallel when write scopes and decisions do not conflict.
5. Use dispatch-protocol.md for parallelization, ownership, failure handling, and merge rules.
6. Use handoff-contracts.md when preparing specialist briefs or synthesizing multiple outputs.
7. Return one unified answer. Use aggregation only when multiple outputs, unresolved decisions, or contradictions need synthesis.

Complexity Levels

• Simple: one narrow domain, one file/component/query/test failure, or ordinary explanation. Handle locally or use one specialist. Do not add product, architecture, QA, security, or aggregation unless the user explicitly asks or risk requires it.
• Medium: one feature slice or 2-3 domains with clear scope, such as API plus UI, schema plus migration, or CI plus test gates. Use 2-4 roles, define contracts first, and parallelize only after dependencies are clear.
• Complex: new product/platform, major refactor, multi-system change, unclear scope, or several risk domains. Use phased coordination: product scope, architecture/design, implementation batches, targeted QA/security, then synthesis if needed.

Common Compositions

Task	Team
Login or account feature	backend-developer + frontend-developer; add security-engineer for auth review
Data-heavy feature	database-engineer for schema/query contract, backend-developer for API, frontend-developer for UI
Small full-stack CRUD	fullstack-developer alone unless separate ownership or risk justifies specialists
Platform build	product-manager, system-architect, then implementation specialists by ownership, then targeted QA/security
CI or deployment	devops-engineer, plus qa-engineer only if quality gates are part of the request

Dispatch Heuristics

• Do not dispatch for a single bug explanation, one SQL query, one component, one file edit, or ordinary Q&A unless the user explicitly asks for team mode.
• Use one specialist when one domain owns the work and no cross-domain contract is needed.
• Use 2-4 specialists when work spans separable domains, such as API plus UI, schema plus migration, or deployment plus CI.

See routing-matrix.md for per-role triggers, avoidance guidance, and the full role catalog.

Parallelism and Handoffs

See dispatch-protocol.md for safe/unsafe parallel patterns, batch building, and conflict handling.

See handoff-contracts.md for specialist brief and aggregation templates.

Compatibility

This skill targets Codex; other agent ecosystems may use the files in agents/ and .claude-plugin/ as role references.