home / skills / rsmdt / the-startup / implementation-planning

implementation-planning skill

/plugins/start/skills/implementation-planning

This skill creates actionable implementation plans that translate features into verifiable tasks, following TDD structure to enable autonomous developer work.

npx playbooks add skill rsmdt/the-startup --skill implementation-planning

Review the files below or copy the command above to add this skill to your agents.

Files (4)
SKILL.md
7.2 KB
---
name: implementation-planning
description: Create and validate implementation plans (PLAN). Use when planning implementation phases, defining tasks, sequencing work, analyzing dependencies, or working on implementation-plan.md files in docs/specs/. Includes TDD phase structure and specification compliance gates.
allowed-tools: Read, Write, Edit, Task, TodoWrite, Grep, Glob
---

# Implementation Plan Skill

Creates actionable implementation plans that break features into executable tasks following TDD principles. Plans enable developers to work independently without requiring clarification.

## Success Criteria

A plan is complete when:
- [ ] A developer can follow it independently without additional context
- [ ] Every task produces a verifiable deliverable (not just an activity)
- [ ] All PRD acceptance criteria map to specific tasks
- [ ] All SDD components have corresponding implementation tasks
- [ ] Dependencies are explicit and no circular dependencies exist

## When to Activate

Activate when:
- **Create a new PLAN** from the template
- **Complete phases** in an existing implementation-plan.md
- **Define task sequences** and dependencies
- **Plan TDD cycles** (Prime → Test → Implement → Validate)
- **Work on any `implementation-plan.md`** file in docs/specs/

## Template

The PLAN template is at [template.md](template.md). Use this structure exactly.

**To write template to spec directory:**
1. Read the template: `plugins/start/skills/implementation-plan/template.md`
2. Write to spec directory: `docs/specs/[NNN]-[name]/implementation-plan.md`

## PLAN Focus Areas

Your plan MUST answer these questions:
- **WHAT** produces value? (deliverables, not activities)
- **IN WHAT ORDER** do tasks execute? (dependencies and sequencing)
- **HOW TO VALIDATE** correctness? (test-first approach)
- **WHERE** is each task specified? (links to PRD/SDD sections)

Keep plans **actionable and focused**:
- Use task descriptions, sequence, and validation criteria
- Omit time estimates—focus on what, not when
- Omit resource assignments—focus on work, not who
- Omit implementation code—the plan guides, implementation follows

## Task Granularity Principle

**Track logical units that produce verifiable outcomes.** The TDD cycle is the execution method, not separate tracked items.

### Good Tracking Units (produces outcome)
- "Payment Entity" → Produces: working entity with tests ✓
- "Stripe Adapter" → Produces: working integration with tests ✓
- "Payment Form Component" → Produces: working UI with tests ✓

### Bad Tracking Units (too granular)
- "Read payment interface contracts" → Preparation, not deliverable
- "Test Payment.validate() rejects negative amounts" → Part of larger outcome
- "Run linting" → Validation step, not deliverable

### Structure Pattern

```markdown
- [ ] **T1.1 Payment Entity** `[activity: domain-modeling]`

  **Prime**: Read payment interface contracts `[ref: SDD/Section 4.2; lines: 145-200]`

  **Test**: Entity validation rejects negative amounts; supports currency conversion; handles refunds

  **Implement**: Create `src/domain/Payment.ts` with validation logic

  **Validate**: Run unit tests, lint, typecheck
```

The checkbox tracks "Payment Entity" as a unit. Prime/Test/Implement/Validate are embedded guidance.

## TDD Phase Structure

Every task follows red-green-refactor within this pattern:

### 1. Prime Context
- Read relevant specification sections
- Understand interfaces and contracts
- Load patterns and examples

### 2. Write Tests (Red)
- Test behavior before implementation
- Reference PRD acceptance criteria
- Cover happy path and edge cases

### 3. Implement (Green)
- Build to pass tests
- Follow SDD architecture
- Use discovered patterns

### 4. Validate (Refactor)
- Run automated tests
- Check code quality (lint, format)
- Verify specification compliance

## Task Metadata

Use these annotations in the plan:

```markdown
- [ ] T1.2.1 [Task description] `[ref: SDD/Section 5; lines: 100-150]` `[activity: backend-api]`
```

| Metadata | Description |
|----------|-------------|
| `[parallel: true]` | Tasks that can run concurrently |
| `[component: name]` | For multi-component features |
| `[ref: doc/section; lines: X-Y]` | Links to specifications |
| `[activity: type]` | Hint for specialist selection |

## Cycle Pattern

For each phase requiring definition, follow this iterative process:

### 1. Discovery Phase
- **Read PRD and SDD** to understand requirements and design
- **Identify activities** needed for each implementation area
- **Launch parallel specialist agents** to investigate:
  - Task sequencing and dependencies
  - Testing strategies
  - Risk assessment
  - Validation approaches

### 2. Documentation Phase
- **Update the PLAN** with task definitions
- **Add specification references** (`[ref: ...]`)
- Focus only on current phase being defined
- Follow template structure exactly

### 3. Review Phase
- **Present task breakdown** to user
- Show dependencies and sequencing
- Highlight parallel opportunities
- **Wait for user confirmation** before next phase

**Ask yourself each cycle:**
1. Have I read the relevant PRD and SDD sections?
2. Do all tasks trace back to specification requirements?
3. Are dependencies between tasks clear?
4. Can parallel tasks actually run in parallel?
5. Are validation steps included in each phase?
6. Have I received user confirmation?

## Specification Compliance

Every phase should include a validation task:

```markdown
- [ ] **T1.3 Phase Validation** `[activity: validate]`

  Run all phase tests, linting, type checking. Verify against SDD patterns and PRD acceptance criteria.
```

For complex phases, validation is embedded in each task's **Validate** step.

### Deviation Protocol

When implementation requires changes from the specification:
1. Document the deviation with clear rationale
2. Obtain approval before proceeding
3. Update SDD when the deviation improves the design
4. Record all deviations in the plan for traceability

## Validation Checklist

See [validation.md](validation.md) for the complete checklist. Key gates:

- [ ] All specification file paths are correct and exist
- [ ] Context priming section is complete
- [ ] All implementation phases are defined
- [ ] Each phase follows TDD: Prime → Test → Implement → Validate
- [ ] Dependencies between phases are clear (no circular dependencies)
- [ ] Parallel work is properly tagged with `[parallel: true]`
- [ ] Activity hints provided for specialist selection `[activity: type]`
- [ ] Every phase references relevant SDD sections
- [ ] Every test references PRD acceptance criteria
- [ ] Integration & E2E tests defined in final phase
- [ ] Project commands match actual project setup
- [ ] A developer could follow this plan independently

## Output Format

After PLAN work, report:

```
📋 PLAN Status: [spec-id]-[name]

Phases Defined:
- Phase 1 [Name]: ✅ Complete (X tasks)
- Phase 2 [Name]: 🔄 In progress
- Phase 3 [Name]: ⏳ Pending

Task Summary:
- Total tasks: [N]
- Parallel groups: [N]
- Dependencies: [List key dependencies]

Specification Coverage:
- PRD requirements mapped: [X/Y]
- SDD components covered: [X/Y]

Next Steps:
- [What needs to happen next]
```

## Examples

See [examples/phase-examples.md](examples/phase-examples.md) for reference.

Overview

This skill creates actionable implementation plans that break features into executable tasks following TDD principles. It produces implementation-plan.md files that map PRD acceptance criteria and SDD components to verifiable deliverables. Plans make dependencies explicit, identify parallel work, and include validation gates for specification compliance.

How this skill works

The skill reads the PLAN template and the relevant PRD/SDD sections, then generates a structured plan using Prime → Test → Implement → Validate steps for each task. Each task is a trackable unit that produces a verifiable outcome, includes specification references and activity metadata, and declares sequencing and parallelism. The final output follows the prescribed template and a PLAN status summary that highlights phases, task counts, and coverage metrics.

When to use it

  • Creating a new implementation-plan.md from the template for a feature or spec
  • Defining or completing phases in an existing implementation-plan.md
  • Sequencing tasks and analyzing dependencies before development starts
  • Planning TDD cycles and embedding validation gates into tasks
  • When you need every PRD acceptance criterion and SDD component mapped to deliverables

Best practices

  • Treat each task as a logical unit that yields a verifiable deliverable, not an activity
  • Always include [ref: ...] links to PRD/SDD sections in Prime steps for traceability
  • Embed Test-first criteria that reference PRD acceptance criteria and edge cases
  • Mark concurrent work with [parallel: true] and ensure no circular dependencies
  • Include a Validate step per task or a Phase Validation task to gate specification compliance

Example use cases

  • Break a new payment feature into entities, adapters, and UI components with tests and validation
  • Define backend API workstreams with clear sequencing, parallel groups, and test-first tasks
  • Convert SDD components into implementation tasks that a developer can follow independently
  • Plan an integration phase that includes E2E tests, lint/type checks, and spec validation
  • Update an implementation-plan.md after a discovery phase to reflect new dependencies

FAQ

How granular should tasks be?

Track logical units that produce outcomes (e.g., Payment Entity), not tiny activities; each must deliver a verifiable artifact and tests.

Should timelines or personnel be included?

No. Plans focus on what to do, order, and validation. Omit time estimates and resource assignments.