ai-team-workflowlisted

## When to Use - Use when: multiple agents need to tackle a feature from different angles simultaneously - Use when: agent output quality improves with a review-before-merge step - Use when: building an autonomous system that can propose and ship without human per-task approval - Do NOT use for: single-agent tasks — overhead not justified - Do NOT use for: real-time coordination — this is async, PR-cycle cadence --- ## Role Taxonomy ``` Architect — breaks down requirements, designs the approach, creates task issues Implementer — writes code from the task spec (one per independent module) Reviewer — audits implementer's work: correctness, tests, security, style QA Agent — runs the test suite, e2e tests, performance checks Integrator — merges approved PRs, resolves conflicts, updates changelog Watchdog — monitors anomalies, triggers escalation on budget overrun or loop ``` --- ## Propose / Vote / Implement / Review / Ship ``` Phase 1 — PROPOSE (Architect agent) → Reads requirements → Proposes 2-3 implementation approaches in a structured doc → Posts to shared channel: .claude/proposals/feat-X.md Phase 2 — VOTE (All senior agents) → Each agent adds vote + rationale to the proposal → Majority wins; tie → escalate to human → Winning approach becomes the task spec Phase 3 — IMPLEMENT (Implementer agents, parallel) → Each implementer claims an independent module → Works in own branch/worktree → Opens PR when complete Phase 4 — REVIEW (Re