sdlc-implementation-plan

name	SDLC-implementation-plan
description	Create a phased implementation plan from design artifacts. Populates tasks.md with short tasks grouped into incremental development phases, each ending with a deployable/testable system. Use after SDLC-decompose, before coding begins.

Instructions

You are working in the Code phase. You are creating an implementation plan — a phased sequence of short development tasks that, when executed in order, build the project incrementally from nothing to full feature set. Each phase ends with a system that can be deployed or at least tested end-to-end, even if not feature-complete.

The plan populates 3-code/tasks.md with tasks and an Execution Plan that groups them into ordered phases.

Phase Validation

Before doing anything else, read the ### Current State subsection under ## Project Overview in CLAUDE.md and determine which phase the project is in. Then follow the matching case below:

1. Project not initialized — the Current State lacks a real project description (e.g., mentions "not yet been initialized" or "base scaffold"). Stop, recommend /SDLC-init, and do not proceed.

2. Project is in the Specification phase — the Current State mentions "Specification phase", lists specification artifacts being drafted, or no phase beyond Specification has been started. Stop, recommend /SDLC-elicit to continue refining specifications or /SDLC-design to start the design phase, and do not proceed.

3. Project is in the Design phase without components identified — Stop, recommend /SDLC-decompose first, and do not proceed.

4. Project is in the Code phase with components identified but no tasks yet — this is the expected entry point. Evaluate prerequisites:

   • (a) All design documents have content — architecture.md, data-model.md, and api-design.md are drafted
   • (b) Components are identified — per-component directories exist in 3-code/
   • (c) 3-code/tasks.md has no tasks yet (only the template)

   Then:

   • If (a) or (b) not met — Stop, explain what is missing.
   • If (c) is not met (tasks already exist) — Warn that running this skill will replace existing task entries and the Execution Plan. Proceed only on explicit user confirmation.
   • If all met — proceed with Setup.

5. Project has tasks already in progress or done — Warn strongly that re-planning may invalidate in-progress work. Proceed only on explicit user confirmation.

Setup

Read the following files — all are required context for planning:

1. Goals — read every GOAL-*.md file in 1-spec/goals/. Extract each goal's title, priority, status, and Success Criteria checkboxes. Only consider goals with Status: Approved.
2. Requirements — read every REQ-*.md file in 1-spec/requirements/. Note their type (Functional, Performance, Security, etc.), priority, and linked goals/stories.
3. Constraints — read every CON-*.md file in 1-spec/constraints/.
4. Design documents — read 2-design/architecture.md, 2-design/data-model.md, and 2-design/api-design.md.
5. Decisions — read every DEC-*.md file in decisions/ (active records only, not .history.md).
6. Code phase instructions — read 3-code/CLAUDE.code.md.
7. Components — read every CLAUDE.component.md in 3-code/*/.
8. Deploy phase instructions — read 4-deploy/CLAUDE.deploy.md.

Analysis

Before constructing the plan, perform these analyses silently (do not output them — they inform the plan):

1. Requirement-to-Component Map

For each requirement, identify which component(s) must implement it (from the component CLAUDE.component.md files). Requirements that span multiple components need tasks in each.

2. Dependency Graph

Identify ordering constraints between requirements (examples for each category):

• Infrastructure dependencies: e.g., database schema before data access, project scaffolding before features, build tooling before components
• Data flow dependencies: e.g., backend API before frontend consumption, authentication service before protected endpoints
• Feature dependencies: e.g., core CRUD before advanced features, basic listing before search/filter

3. Priority Ordering

Group requirements by goal priority:

• Must-have goals first — their requirements form the earliest phases
• Should-have goals later — their requirements come after Must-have coverage
• Could-have goals last

4. Cross-Cutting Concerns

Identify work that does not map to a single requirement but is needed for the system to function:

• Project scaffolding (package.json, pyproject.toml, directory structure, virtual environments)
• Build and dev tooling (dev server, hot reload, linters)
• Database initialization (schema creation, migrations)
• Integration wiring (backend serves frontend, CORS, static files)
• Infrastructure as code (Terraform, AWS SAM, Dockerfiles)
• Operational runbooks (startup procedures, deployment scripts, troubleshooting guides)

Plan Construction

Build the plan following these principles:

Principle 1: Each Phase Produces a Working System

Every phase must end with a system that can be started and tested, even if incomplete. This means:

• Phase 1 always includes project scaffolding, build tooling, and a minimal running application
• Each subsequent phase adds capabilities while keeping the system functional
• No phase should leave the system in a broken state

Principle 2: Incremental Capability Delivery

Each phase delivers a coherent set of new capabilities — features that were not available in the previous phase. A capability maps to one or more Success Criteria checkboxes from the goals.

Define each phase with:

• Name: a short descriptive name (e.g., "Project Scaffolding & Dev Environment", "Core Data Pipeline", "User Dashboard")
• Capabilities delivered: a bulleted list of what becomes possible after this phase that was not possible before. Reference goal Success Criteria where applicable.
• Tasks: the ordered list of short tasks within this phase

Principle 3: Short Tasks

Each task should represent a small, focused unit of work — ideally completable by human developers in a single coding session. Guidelines:

• A task implements one feature, one endpoint, one component, or one piece of infrastructure
• If a task description contains "and" connecting two distinct pieces of work, split it
• Prefer many small tasks over few large ones
• Ideally, each task should be individually testable or verifiable

Principle 4: Must-Haves Before Should-Haves

Phases that deliver Must-have goal capabilities come before phases that deliver Should-have capabilities. Within a priority level, order by dependency graph.

Principle 5: Infrastructure and Integration Early

Cross-cutting infrastructure (scaffolding, build tooling, database schema, deployment configuration, unit and integration testing pipelines) should appear in early phases so that feature work can build on a solid foundation.

Principle 6: Deploy Readiness

If 4-deploy/ expectations exist, include infrastructure-as-code tasks (Terraform modules, SAM templates, Dockerfiles, deployment scripts) and operational runbook tasks at appropriate phases — not all at the end. A basic deployment mechanism should be available relatively early so the system can be tested in a realistic environment.

Principle 7: Manual Testing Readiness

Every phase must end with a manual testing readiness task (TASK-phase-N-manual-testing) that ensures a human can verify the delivered capabilities without reading source code. The task should:

1. Create or update a deploy runbook in 4-deploy/runbooks/ with startup instructions and step-by-step manual test scenarios for the phase's new capabilities. Update existing runbooks rather than creating duplicates.
2. Document build/run/test commands in the README (or equivalent existing doc) of every component modified in this phase or previous phases (e.g., 3-code/<component>/README.md). Create or update convenience scripts where useful; for single-command operations, documenting the command is sufficient.

The Component column must list all components involved in the current phase and previous phases. Place these tasks in the ### Deploy & Operations task table section, dependent on the phase's last functional task, with - in the Req column. List them last in the Execution Plan for their phase.

Workflow

1. Present the Implementation Plan

Present the complete plan to the user in this format:

## Implementation Plan

### Phase 1: [Name]

**Capabilities delivered:**
- [What becomes possible after this phase]
- [Reference GOAL-* Success Criteria where applicable]

**Tasks:**

| # | Task ID | Task | Component | Req |
|---|---------|------|-----------|-----|
| 1 | TASK-kebab-name | Short description | backend | REQ-F-* |
| 2 | TASK-kebab-name | Short description | frontend | REQ-F-* |
| 3 | TASK-phase-1-manual-testing | Create/update runbook and scripts for manual testing of phase capabilities | backend, frontend | - |

### Phase 2: [Name]
...

For each phase, explain briefly why these capabilities are grouped together and why this ordering makes sense.

After the plan, provide a summary table:

Phase	Name	# Tasks	Goal Coverage
1	...	N	GOAL-a, GOAL-b
2	...	N	GOAL-c
...	...	...	...

And a coverage check: list each approved goal and confirm which phase(s) address its Success Criteria. Flag any Success Criteria not covered by any task.

Wait for user approval before proceeding. The user may reorder phases, split or merge them, add or remove tasks, or change priorities.

2. Populate tasks.md

After approval, write the approved plan into 3-code/tasks.md:

Task Table sections: Create one section per component (matching the per-component directories in 3-code/) plus ### Setup & Infrastructure for cross-cutting tasks and ### Deploy & Operations for IaC and runbooks. Place each task in the section matching its component.

For each task, create a row:

ID	Task	Priority	Status	Req	Dependencies	Updated	Notes

• ID: TASK-kebab-name — descriptive, unique
• Task: short description (imperative form)
• Priority: derive from the goal priority (Must-have goal → P1, Should-have → P2, infrastructure/cross-cutting → P0 or P1 based on criticality)
• Status: Todo
• Req: link(s) to requirement file(s) this task implements, or - for infrastructure tasks. Example: [REQ-F-search-by-name](../1-spec/requirements/REQ-F-search-by-name.md)
• Dependencies: TASK-* IDs this task depends on, or -
• Updated: today's date (YYYY-MM-DD)
• Notes: empty or brief clarification

Execution Plan section: Replace the placeholder blocks with the approved phases:

### Phase 1: [Name]

**Capabilities delivered:**
- [capability]

**Tasks:**
1. TASK-kebab-name
2. TASK-kebab-name

### Phase 2: [Name]
...

3. Update Current State

After writing tasks, update the ### Current State subsection in CLAUDE.md to record that the implementation plan has been created. Include:

• Number of phases
• Total number of tasks
• Date of plan creation

Example addition: "Implementation plan created (2026-03-12): N phases, M tasks covering all approved goals."

Interaction Style

• Present the full plan at once — do not drip-feed phases one at a time.
• Be explicit about trade-offs: if a capability could go in phase N or N+1, explain why you placed it where you did.
• If a goal's Success Criteria cannot be fully covered by implementation tasks alone (e.g., requires manual verification, external resources), note this in the coverage check.
• After writing tasks.md, summarize what was written and ask if adjustments are needed.

Rules

• Read-before-write: always read existing files before proposing changes.
• Do not create source code: this skill creates the plan only. Actual implementation happens when tasks are executed.
• Do not modify design documents: if a design gap is discovered during planning, surface it to the user and recommend running /SDLC-design to address it.
• Do not modify objective artifacts: if a requirement gap is discovered, surface it and recommend /SDLC-elicit.
• Preserve existing content: when updating tasks.md, preserve the Status Legend, Priority Legend, and How to Update sections. Replace only the Task Table sections and Execution Plan section.
• User approval required: present the full plan and wait for explicit approval before writing to tasks.md.
• Task ID uniqueness: every TASK-* ID must be unique. Use descriptive kebab-case names, not numbers.
• Traceability: every task implementing a requirement must link to it in the Req column. Tasks not linked to requirements (infrastructure, tooling) should use -.