Budget Analyser — Team Collaboration Model

Teams

Planning Team

Implementation Team

Quality Team

Cross-cutting

Collaboration Protocol

Phase 1: Round-table Discussion

When a new task arrives, the Software Architect initiates a round-table:

New Task
  │
  ├── Software Architect    → Proposes design approach, identifies components
  ├── Finance Analyst       → Validates domain logic, flags financial edge cases
  ├── Frontend Engineer     → Flags UI complexity, estimates frontend effort
  ├── Backend Engineer      → Flags data/API complexity, estimates backend effort
  │
  └── Consensus → Brief design decision + task breakdown

Participation rules:

• Software Architect: ALWAYS participates
• Finance Analyst: Joins when task touches financial logic, categories, budgets, reports, forecasting, or transaction data
• UI/UX Engineer: Joins when task involves layout, design, accessibility, or financial data presentation
• Frontend Engineer: Joins when task involves UI changes or new pages
• Backend Engineer: Joins when task involves API, service, model, or data changes
• Pytest/Playwright Engineers: Join when task has significant testability implications

Round-table output: Design decision document with:

• Agreed approach and rationale
• Task breakdown per agent
• Identified risks and mitigations
• Dependencies between tasks

Phase 2: Parallel Dispatch

After consensus, implementation agents work simultaneously:

Design Decision
  │
  ├── UI/UX Engineer    ──→ Layout specs, component design, accessibility (first)
  │
  ├── Frontend Engineer ──→ React components, pages, hooks (worktree A)
  │
  ├── Backend Engineer  ──→ Vertical slice: models.py + service.py + router (worktree B)
  │
  └── FE + BE work independently, merge when complete

Rules:

• Use isolated git worktrees when both agents modify code
• Each agent follows the approved design — no freelancing
• Flag deviations immediately rather than proceeding with unapproved changes

Phase 3: Quality Pipeline

Sequential validation after implementation:

Implementation Complete
  │
  ├── 1. Pytest Engineer     → Unit tests + integration tests
  │                            Run: uv run pytest src/test/unit/ -q
  │
  ├── 2. Playwright Engineer → E2E browser tests
  │                            Run: cd src/frontend && npm run test:e2e
  │
  └── 3. Code Reviewer       → Final review gate
                                Verdict: APPROVE or REQUEST CHANGES

Quality gate rules:

• All unit tests must pass before E2E testing begins
• Code Reviewer review is MANDATORY for every implementation task
• REQUEST CHANGES blocks merge until issues are resolved

Mandatory Review Matrix

Universal Standards

All agents MUST:

1. Read CLAUDE.md before starting any work
2. Follow vertical slices architecture pattern
3. Write Google-style docstrings on all public functions and classes
4. Add type hints on all function signatures
5. Use from __future__ import annotations in every Python module
6. Create signed semantic commits with file-change tables
7. Respect max line length of 100 characters
8. Use keyword-only arguments for clarity (def foo(*, arg1, arg2))

Agent Roster