Claude Code Plugins

Community-maintained marketplace

Feedback

Create technical implementation plans from specifications with intelligence-backed architectural decisions. Use when specification exists and user mentions tech stack, architecture, implementation approach, or asks "how to implement" or "how to build" the feature.

Install Skill

1Download skill
2Enable skills in Claude

Open claude.ai/settings/capabilities and find the "Skills" section

3Upload to Claude

Click "Upload skill" and select the downloaded ZIP file

Note: Please verify skill by going through its instructions before using it.

SKILL.md

name Implementation Planning
description Create technical implementation plans from specifications with intelligence-backed architectural decisions. Use when specification exists and user mentions tech stack, architecture, implementation approach, or asks "how to implement" or "how to build" the feature.
degree-of-freedom medium
allowed-tools Bash(project-intel.mjs:*), Read, Write, Edit

@.claude/shared-imports/constitution.md @.claude/shared-imports/CoD_Σ.md @.claude/templates/plan.md

Implementation Planning

Purpose: Transform technology-agnostic specifications into technical implementation plans with architecture decisions, tech stack selection, and acceptance criteria mapping.

Constitutional Authority: Article IV (Specification-First Development), Article VI (Simplicity & Anti-Abstraction), Article I (Intelligence-First Principle)


Phase 0: Pre-Design Constitutional Gates

MANDATORY: Check Article VI limits BEFORE design.

Step 0.1: Validate Specification Exists

PreToolUse hook will block plan.md creation without spec.md, but verify explicitly:

FEATURE=$(git rev-parse --abbrev-ref HEAD 2>/dev/null || echo "${SPECIFY_FEATURE:-}")
if [[ ! -f "specs/$FEATURE/spec.md" ]]; then
    echo "ERROR: Cannot create plan without specification (Article IV violation)"
    echo "Next: Run specify-feature skill or use /feature command"
    exit 1
fi

Step 0.2: Constitution Check (Article VI)

Read specification and assess complexity:

Gate 1: Project Count

  • Maximum 3 projects for initial implementation
  • If spec implies 4th project → require justification

Gate 2: Abstraction Layers

  • Maximum 2 abstraction layers per concept
  • No repository/service/facade patterns without documented need

Gate 3: Framework Trust

  • Use framework features directly (no wrappers)
  • Justify any custom implementations

If violations detected, create Complexity Justification section:

Violation Why Needed Simpler Alternative Rejected Because
[e.g., 4th project] [reason] [why 3 projects insufficient]

Phase 1: Intelligence-First Context Gathering

Article I Mandate: Query before reading files.

Step 1.1: Search for Existing Patterns

!`project-intel.mjs --search "<tech-stack-keywords>" --type "tsx,ts,py,go" --json > /tmp/plan_intel_patterns.json`

Example:

  • If spec mentions "authentication" → search for "auth login session"
  • If spec mentions "real-time" → search for "websocket socket io"

Save evidence to /tmp/plan_intel_patterns.json

Step 1.2: Analyze Project Architecture

!`project-intel.mjs --overview --json > /tmp/plan_intel_overview.json`

Understand:

  • Existing tech stack and frameworks
  • Directory structure patterns
  • Naming conventions
  • Test frameworks in use

Step 1.3: Query Dependencies and Integration Points

!`project-intel.mjs --dependencies "<key-files>" --direction both --json > /tmp/plan_intel_deps.json`

Identify integration points where new feature connects to existing code.


Phase 2: Technical Design

Step 2.1: Select Tech Stack

Based on intelligence findings and spec requirements:

Language/Version: [from project-intel overview or specify new] Primary Dependencies: [frameworks, libraries with versions] Storage: [database, caching, file system if applicable] Testing: [test frameworks, assertion libraries] Target Platform: [web, mobile, desktop, server]

CoD^Σ Evidence:

Tech Stack Selection:
project-intel.mjs --overview → existing stack is TypeScript + React + Supabase
Decision: Continue with existing stack for consistency
Evidence: src/*/package.json:12 shows react@18, typescript@5

Step 2.2: Design Architecture

Component Breakdown (from spec user stories):

For each P1 user story:

  1. Identify required components (models, services, UI, API)
  2. Map to existing architecture (from intelligence queries)
  3. Document new components needed
  4. Define interfaces/contracts

Example:

User Story P1: Email/Password Authentication

Components:
- Model: User (extends existing User model at models/user.ts:8)
- Service: AuthService (new, will integrate with existing auth.ts:23)
- API: POST /auth/login, POST /auth/register (new endpoints)
- UI: LoginForm, RegisterForm (enhance existing at components/auth/)

Integration Points:
- Existing: src/utils/auth.ts:23 (session management)
- New: src/services/auth-service.ts (authentication logic)

Step 2.3: Create Research Document

Generate research.md with:

Technical Decisions:

  1. Decision: [what was decided]
    • Rationale: [why]
    • Alternatives considered: [what else was considered]
    • Evidence: [file:line from intelligence or MCP query]

Example:

## Decision 1: OAuth Provider Libraries

Decision: Use @supabase/auth-helpers for OAuth integration

Rationale:
- Already using Supabase for backend (src/lib/supabase.ts:5)
- Native support for Google/GitHub OAuth
- Handles token refresh automatically

Alternatives Considered:
- NextAuth.js: Rejected (adds complexity, not needed with Supabase)
- Custom OAuth: Rejected (reinventing wheel, Article VI violation)

Evidence:
- Intelligence: project-intel found Supabase client at src/lib/supabase.ts:5
- MCP Ref: @supabase/auth-helpers supports Google/GitHub OAuth

Step 2.4: Design Data Model

Generate data-model.md (entities WITHOUT implementation):

For each entity:

  • Name and purpose
  • Attributes (without database types)
  • Relationships to other entities
  • Validation rules

Example:

## Entity: User

Purpose: Represents authenticated system user

Attributes:
- id: Unique identifier
- email: Email address (unique, validated)
- password_hash: Hashed password (never plaintext)
- oauth_provider: Optional (google|github)
- oauth_id: Optional provider-specific ID
- created_at: Registration timestamp
- last_login: Last successful login

Relationships:
- Has many: Sessions
- Has many: UserRoles

Validation:
- email: Must be valid email format
- password: Minimum 8 characters, must include number + special char
- oauth_provider + oauth_id: Required together or both null

Step 2.5: Define API Contracts

Generate contracts/ directory with API specifications:

For each endpoint:

  • HTTP method and path
  • Request schema
  • Response schema
  • Error cases
  • Authentication required?

Example: contracts/auth-endpoints.md

## POST /api/auth/register

Purpose: Create new user account

Request:
- email: string (required, valid email)
- password: string (required, min 8 chars)

Response (201 Created):
- user: { id, email, created_at }
- session: { access_token, refresh_token, expires_at }

Errors:
- 400: Invalid email format
- 409: Email already exists
- 500: Server error

Authentication: None (public endpoint)

Step 2.6: Create Quickstart Validation Scenarios

Generate quickstart.md with test scenarios:

For each user story, document:

  • Setup steps
  • Exact actions to test
  • Expected outcomes
  • How to verify success

Example:

## Scenario 1: User Registration (P1 Story)

Setup:
1. Navigate to /register
2. Have valid email ready

Test Steps:
1. Enter email: test@example.com
2. Enter password: SecurePass123!
3. Click "Register"

Expected Outcome:
- HTTP 201 response
- User created in database
- Session token returned
- Redirected to /dashboard

Verification:
- Check database: SELECT * FROM users WHERE email='test@example.com'
- Check session storage: localStorage.getItem('session')
- Verify redirect: window.location.pathname === '/dashboard'

Phase 3: Acceptance Criteria Mapping

Article III Requirement: Minimum 2 testable ACs per user story.

Step 3.1: Extract User Stories from Specification

Read specs/$FEATURE/spec.md

Identify all user stories with priorities (P1, P2, P3).

Step 3.2: Generate Acceptance Criteria

For EACH user story, create ≥2 ACs:

AC Format:

  • AC-ID: [Story]-[Number] (e.g., P1-001, P1-002)
  • Given [precondition], When [action], Then [outcome]
  • Test: How to automate verification

Example:

### User Story P1: Email/Password Registration

**AC-P1-001**: User can register with valid email and password
- **Given** user has valid email and strong password
- **When** user submits registration form
- **Then** account is created and user is logged in
- **Test**: POST /api/auth/register with valid data returns 201 + session token

**AC-P1-002**: System rejects weak passwords
- **Given** user has valid email but weak password ("12345")
- **When** user submits registration form
- **Then** registration fails with error "Password must be at least 8 characters with number and special character"
- **Test**: POST /api/auth/register with weak password returns 400 with specific error

Phase 4: Post-Design Constitutional Re-Check

Article VI Mandate: Verify gates still pass after design.

Step 4.1: Re-validate Complexity

Check if design introduced violations:

  • Did component breakdown create 4th project?
  • Did architecture add unnecessary abstraction layers?
  • Did design create wrappers around framework features?

If new violations: Update Complexity Justification table.

Step 4.2: Verify Framework Trust

Confirm design uses framework features directly:

  • Authentication: Using Supabase Auth (not custom)
  • Database: Using Supabase client (not ORM wrapper)
  • State management: Using framework state (not custom abstraction)

Phase 5: Generate Implementation Plan

Step 5.1: Write plan.md

Use @.claude/templates/plan.md structure:

---
feature: <number>-<name>
created: <YYYY-MM-DD>
specification: specs/<number>-<name>/spec.md
status: Ready for Implementation
---

Sections:

  1. Summary: One paragraph overview
  2. Technical Context: Tech stack, platform, dependencies
  3. Constitution Check: Gates passed/violated with justifications
  4. Architecture: Component breakdown with integration points
  5. Acceptance Criteria: All ACs from Step 3.2
  6. File Structure: Exact paths where code will live
  7. CoD^Σ Evidence: Intelligence query results, MCP sources

Save:

  • specs/$FEATURE/plan.md
  • specs/$FEATURE/research.md
  • specs/$FEATURE/data-model.md
  • specs/$FEATURE/contracts/*.md
  • specs/$FEATURE/quickstart.md

Step 5.2: Report Completion

Output:

✓ Implementation plan created: specs/<number>-<name>/plan.md

Intelligence Evidence:
- project-intel.mjs queries: 3 executed, results in /tmp/plan_intel_*.json
- Existing patterns found: auth.ts:23, Login.tsx:12
- Integration points: 2 identified

Technical Decisions:
- Tech stack: TypeScript + React + Supabase (existing)
- New components: AuthService, LoginForm, RegisterForm
- Architecture: Extends existing auth pattern

Artifacts Generated:
- plan.md (implementation plan)
- research.md (technical decisions)
- data-model.md (User entity)
- contracts/auth-endpoints.md (API specs)
- quickstart.md (test scenarios)

Acceptance Criteria: 6 total (2 per P1 story)

Constitutional Compliance:
- ✓ Article I: Intelligence queries executed before design
- ✓ Article IV: Plan references existing spec.md
- ✓ Article VI: No complexity violations (3 projects, framework trust maintained)
- ✓ Article III: All user stories have ≥2 ACs

**Automatic Task Generation**:
Now invoking generate-tasks skill to create task breakdown...

[Task generation and audit will proceed automatically]

Phase 6: Automatic Task Generation

DO NOT ask user to trigger task generation - this is automatic workflow progression.

Step 6.1: Invoke generate-tasks Skill

After all plan artifacts are complete (plan.md, research.md, data-model.md, contracts/, quickstart.md), automatically proceed to task generation.

Instruct Claude:

"Now that the implementation plan is complete, automatically generate the task breakdown:

Use the generate-tasks skill to create tasks.md.

This will:

  1. Load user stories from specs/$FEATURE/spec.md
  2. Load technical details from specs/$FEATURE/plan.md
  3. Organize tasks by user story (Article VII)
  4. Mark parallelizable tasks with [P]
  5. Ensure ≥2 ACs per story (Article III)
  6. Automatically invoke /audit for quality gate validation

The generate-tasks skill will handle this entire workflow automatically, including the quality gate check."

Step 6.2: Quality Gate Enforcement

The generate-tasks skill will automatically invoke /audit after creating tasks.md.

This enforces:

  • Article IV: Specification-First Development (spec → plan → tasks validated)
  • Article V: Template-Driven Quality (automatic validation)
  • Article VII: User-Story-Centric Organization (verified)

User will see:

  • tasks.md created
  • /audit runs automatically
  • Quality gate PASS/FAIL result
  • Next steps based on audit outcome

No manual intervention required - workflow proceeds automatically through task generation and validation.


Note: This completes the planning phase. The next steps (task generation + audit) happen automatically.


---

## Anti-Patterns to Avoid

**DO NOT**:
- Plan before specification exists (Article IV violation)
- Skip intelligence queries (Article I violation)
- Design without checking constitution gates (Article VI violation)
- Create ACs with < 2 per story (Article III violation)
- Mix specification and implementation concerns
- Copy existing code without intelligence analysis
- Create wrapper layers around framework features

**DO**:
- Query intelligence sources before designing
- Check constitution gates pre AND post design
- Create ≥2 testable ACs per user story
- Use CoD^Σ traces with file:line evidence
- Trust framework features (avoid custom implementations)
- Document decisions with rationale in research.md
- Map components to existing architecture patterns

---

**Skill Version**: 1.0.0
**Last Updated**: 2025-10-19