Subagent Driven Development

Sequential (one at a time):

• Tasks depend on each other
• Same files being modified
• Need to see results before next step

Parallel (simultaneous):

• Tasks are independent (different files/subsystems)
• No shared state
• Can be reviewed together at the end

3. Dispatch Subagents

For each task, dispatch a fresh agent:

Task tool:
  subagent_type: [ui-engineer | senior-backend-engineer | general-purpose]
  description: "Implement [task name]"
  prompt: |
    Task: [clear description]

    Requirements:
    - [specific requirement 1]
    - [specific requirement 2]

    Constraints:
    - Read AGENTS.md for project rules
    - Use cargo test / cargo clippy
    - Follow TDD

    Report back: What you implemented, test results, files changed

4. Review After Each Task (or Batch)

After subagent completes:

Task tool:
  subagent_type: mtg-code-reviewer
  description: "Review [task name] implementation"
  prompt: |
    Review changes for [task name].

    Check:
    1. Code quality
    2. Test coverage
    3. AGENTS.md compliance

    Return: Issues (Critical/Important/Minor), Assessment

If Critical issues: Fix before proceeding If Important issues: Fix before next task If Minor issues: Note for later

5. Finalize

After all tasks complete:

• Run full test suite: cargo test
• Use /git-workflow to commit

Example: UI + Backend in Parallel

Task 1 (parallel): ui-engineer
  "Implement HandDisplay Bevy plugin showing cards in player's hand"

Task 2 (parallel): senior-backend-engineer
  "Add get_hand query to return player's hand from GameSnapshot"

[Both complete]

Task 3: mtg-code-reviewer
  "Review HandDisplay and get_hand implementations"

[Fix any issues]

/git-workflow

Red Flags

Stop if:

• Subagent is stuck or confused -> provide more context
• Tasks are conflicting -> make sequential instead
• Scope is growing -> pause and re-plan

Never:

• Skip code review
• Proceed with Critical issues unfixed
• Let parallel agents edit same files