MasterMind: Breakdown

tasks.md is the build order: small enough that each task is checkable, ordered so each user story ships alone. Layout and IDs: ~/.mastermind/engineering/core/spec-driven.md. Input: $ARGUMENTS.

Read first

spec.md for the stories and their priorities, plan.md for paths and structure, then data-model.md, contracts/, research.md and quickstart.md where they exist. Use specs/.templates/tasks.md if present, else template.md.

The task line

- [ ] T014 [P] [US1] Add the export button to src/reports/ReportHeader.tsx (US1/AC1, FR-002)

• - [ ] then a T### ID in execution order. IDs are never reused or renumbered.
• [P] only when it touches different files from every unfinished task before it.
• [US#] on every task inside a story phase, and on no other.
• The exact file path, and the spec IDs it serves in parentheses.

The phases

1. Setup: what the project needs before anything else, such as dependencies, config and env vars.
2. Foundation: what every story needs and no story can deliver alone. Keep it small; most work belongs to a story.
3. One phase per user story, in priority order. Each is a vertical slice: its models, logic, interfaces and wiring, ending in a checkpoint that runs the story's independent test.
4. Polish: cross-cutting work that touches several stories.

Delete a phase with no tasks. Never write a heading over "None".

Refactoring the code needs before a story can land goes first, as its own tasks in Foundation. A change too wide for one slice, such as renaming a column every story reads, is sequenced expand, migrate, contract (deprecate), never forced into a story.

Within a story: tests first when they are required, then data, then logic, then the surface, then the wiring that connects them.

Rules

• Tests are tasks only when required. Required means the constitution demands them, the spec asks for them, or the user said so. Otherwise qa verifies, and tests are offered after the build.
• Quote the constraint. A field with a length limit, an enum or a validation rule from data-model.md has that rule copied into its task, so it is not left to memory at build time.
• Every requirement gets a task. Before writing the file, check each FR-### and each acceptance criterion maps to at least one task. List any that do not in your report.
• A task that needs "and" is two tasks. One task, one change, one way to tell it worked.

Gotchas

• Horizontal phases are the failure this prevents. "All models, then all services, then all UI" hides every integration bug until the end. Group by story, never by layer.
• Do not plan here. A task that introduces a library, file or pattern the plan does not mention means the plan is incomplete. Fix plan.md, then the tasks.
• Regenerating loses progress. When a tasks.md with ticked boxes exists, add and amend tasks; never rewrite the file from scratch.

Report

The task count per phase, the parallel opportunities, the requirements with no task, and the next step: analyze before building a feature that touches money, auth or data, else build. Set the spec's status to tasked.

Then run mastermind next and end with the step it names, so the user sees where the feature stands.