Smirnov Labs Claude Skills

A Claude Code skills marketplace and starter template that bundles reusable AI workflows, a Memory Bank system for persistent project context, custom slash commands, and an automated architecture reviewer -- everything you need to supercharge Claude Code on any project.

Quick Start: Install Skills

# 1. Register this marketplace (one-time)
/plugin marketplace add smirnov-labs/claude-skills

# 2. Install any plugin (pick the ones you want)
/plugin install presentation-tools@smirnovlabs-claude-skills
/plugin install plan-utilities@smirnovlabs-claude-skills
/plugin install git-recon@smirnovlabs-claude-skills
/plugin install pr-utilities@smirnovlabs-claude-skills

Or from the CLI outside a session:

claude plugin marketplace add smirnov-labs/claude-skills
claude plugin install presentation-tools@smirnovlabs-claude-skills

Once installed, skills activate automatically based on what you ask. Say "build a pitch deck for this project", "get a second opinion on this plan", or "run a git recon on this repo" — or invoke directly with /presentation-blueprint, /codex-plan-review, or /git-recon.

Available Skills

| Skill | Plugin | What it does | Trigger phrases | |-------|--------|--------------|-----------------| | presentation-blueprint | presentation-tools | End-to-end presentation consultant: analyzes codebases/websites/projects, crafts strategic blueprints, and renders polished decks via document-skills:pptx | "build a deck", "create a presentation", "pitch deck for" | | codex-plan-review | plan-utilities | Sends your current plan to Codex 5.4 (high-effort, read-only) for an independent architectural second opinion before you execute | "review plan with codex", "get a second opinion", "iterate on plan" | | detect-overengineering | plan-utilities | Audits an implementation-plan doc for LLM-style overengineering: a rubric-driven Claude pass plus a Codex adversarial pass, merged into specific edits that slim the plan | "check this plan for bloat", "is this overengineered", "detect overengineering" | | git-recon | git-recon | 12-month git-history health check: surfaces high-churn files, bus factor, bug hotspots, velocity trends, and firefighting patterns before you read any code | "git recon", "codebase health check", "where are the landmines" | | reading-bot-pr-reviews | pr-utilities | Read-only: fetches every automated PR review (Codex across its 3 endpoints + the Architecture/SAR review), filters on exact bot logins, and reports a severity-classified merge verdict so no finding — or clean pass — is missed | "what did the bots say", "read the PR reviews", "did Codex/the SAR pass" |

Prerequisites

The presentation-blueprint skill requires document-skills:pptx for rendering:

/plugin install document-skills@anthropic-agent-skills

The codex-plan-review skill requires the Codex CLI on your PATH with a gpt-5.4-codex model available.

The git-recon skill needs only git — runs locally against the repo you're standing in.

The reading-bot-pr-reviews skill requires the GitHub CLI (gh) authenticated against the repo, and a repo whose PRs are reviewed by Codex and/or an Architecture-Review GitHub Action.

What's in This Repo

| Component | Location | Description | |-----------|----------|-------------| | Skills Marketplace | .claude-plugin/marketplace.json + skills/ | Installable skills that other Claude Code instances can pull from github:smirnov-labs/claude-skills | | Memory Bank | .claude/memory-bank/ | Structured documentation templates that give Claude full project context across sessions | | Custom Slash Commands | .claude/commands/ | /plan, /pr, /memory, /loop-review -- ready-to-use workflows | | Architect-Gate | .github/claude/ | Automated architecture review that runs on every PR via CI |

Quick Start: Install Skills
Using as a Project Template
What's Included
How It Works
Usage Guide
Memory Bank Files
Best Practices
Advanced Usage
Adding Your Own Skills
Troubleshooting
Contributing

Using as a Project Template

Want the full Memory Bank + Architect-Gate + commands for a new project? Clone the repo:

git clone https://github.com/smirnov-labs/claude-skills.git my-project
cd my-project
rm -rf .git  # Remove template git history
git init     # Start fresh

Then fill out .claude/memory-bank/projectbrief.md with your project details -- this is your foundation. Work through the other template files (productContext.md, systemPatterns.md, techContext.md, activeContext.md, progress.md) as your project takes shape.

Why a Memory Bank?

Claude Code's memory resets between sessions. The Memory Bank solves this:

Structured Documentation: Pre-built templates for capturing project context
Automatic Context Loading: Claude reads your memory bank at every session start
Planning Workflows: Built-in /plan command for structured feature development
Project Intelligence: System that learns and captures project-specific patterns

What's Included

.claude/
├── claude.md                    # Main configuration & project intelligence
├── commands/
│   ├── plan.md                  # /plan — structured feature planning
│   ├── pr.md                    # /pr — pull request workflow
│   ├── memory.md                # /memory — memory bank management
│   └── loop-review.md           # /loop-review — iterative plan refinement
└── memory-bank/
    ├── QUICKSTART.md            # Quick start guide
    ├── memory-rules.md          # Complete system documentation
    ├── projectbrief.md          # Foundation document (template)
    ├── productContext.md        # Product vision (template)
    ├── systemPatterns.md        # Architecture patterns (template)
    ├── techContext.md           # Tech stack & setup (template)
    ├── activeContext.md         # Current work focus (template)
    └── progress.md              # Status tracking (template)

.claude-plugin/
└── marketplace.json             # Skills marketplace registry

.github/claude/
├── architect-gate.yml           # CI workflow for architecture review
└── ...                          # Review prompts and configuration

skills/
├── presentation-blueprint/      # Presentation consultant skill
│   ├── SKILL.md                 # Skill definition
│   ├── narrative-frameworks.md  # Reference: storytelling structures
│   ├── slide-archetypes.md      # Reference: slide design patterns
│   └── source-analysis.md       # Reference: source evaluation guide
├── codex-plan-review/           # Codex second-opinion skill
│   └── SKILL.md                 # Skill definition
├── detect-overengineering/      # Plan overengineering audit skill
│   └── SKILL.md                 # Skill definition
├── git-recon/                   # Git history health-check skill
│   └── SKILL.md                 # Skill definition
└── reading-bot-pr-reviews/      # Automated PR-review reader skill
    └── SKILL.md                 # Skill definition

Key Components

1. Skills Marketplace (.claude-plugin/ + skills/)

Installable via github:smirnov-labs/claude-skills
Ships with four plugins: presentation-tools, plan-utilities, git-recon, and pr-utilities

2. Slash Commands (.claude/commands/)

/plan -- structured feature planning with clarifying questions
/pr -- pull request creation workflow
/memory -- memory bank review and update
/loop-review -- iterative plan refinement loop

3. Memory Bank (.claude/memory-bank/)

6 core template files for project documentation
Complete system documentation
Quick start guide

4. Architect-Gate (.github/claude/)

Automated architecture review on every PR
Anti-pattern detection and severity scoring
Posts a System Architecture Review (SAR) comment

5. Configuration (.claude/claude.md)

Automatic memory loading via alwaysApply: true
Project intelligence journal
Instructions for Claude on how to use the system

How It Works

Automatic Context Loading

The .claude/claude.md file has alwaysApply: true configured, which means:

1. Every Session Start: Claude automatically reads the instructions in claude.md 2. Memory Bank Loading: Instructions tell Claude to read all memory bank files in order 3. Immediate Context: Claude has full project understanding without you explaining anything

File Hierarchy

Files build upon each other logically:

projectbrief.md (Foundation)
    ├── productContext.md (Why it exists)
    ├── systemPatterns.md (How it's built)
    └── techContext.md (What technologies)
        └── activeContext.md (Current state - synthesizes all)
            └── progress.md (Tracks active context over time)

Update Workflow

When you've made significant changes:

update memory bank

Claude will: 1. Review ALL memory bank files (mandatory) 2. Update files that need changes 3. Focus on activeContext.md and progress.md 4. Document new patterns in claude.md

Usage Guide

Starting a New Session

Just open Claude Code - memory loads automatically:

claude

Claude now has full context. Start working immediately:

Let's continue with the authentication feature

Planning a Feature

Use structured planning for complex work:

/plan

I want to add user authentication with OAuth

Claude will:

Review memory bank
Ask clarifying questions (OAuth provider? Token storage? Session management?)
Draft comprehensive plan
Get your approval
Execute systematically

Daily Workflow

1. Start Session: Open Claude Code (context loads automatically) 2. Check Context: Claude references activeContext.md for current priorities 3. Work: Implement features, fix bugs, refactor 4. Update: Say "update memory bank" after significant changes 5. Plan Next: Use /plan for next major feature

Documenting as You Go

Update relevant files during development:

Making architectural decisions? → Update systemPatterns.md
Adding dependencies? → Update techContext.md
Completing features? → Update progress.md
Switching focus? → Update activeContext.md

Say "update memory bank" and Claude will help you document.

Memory Bank Files

projectbrief.md

Purpose: Foundation document

Update Frequency: Rarely (only when project scope changes)

Contains: Goals, scope, constraints, success criteria

This is your source of truth. All other files derive from it.

productContext.md

Purpose: Product vision and user experience

Update Frequency: Occasionally (when product direction shifts)

Contains: Problems solved, user workflows, requirements

Answers "why does this exist?" and "who is it for?"

systemPatterns.md

Purpose: Architecture and technical patterns

Update Frequency: Regularly (as patterns emerge)

Contains: Architecture, design patterns, technical decisions, critical paths

Include code examples and file references with line numbers.

techContext.md

Purpose: Technology stack and setup

Update Frequency: When adding/removing dependencies

Contains: Tech stack, dependencies, setup instructions, infrastructure

Your technical reference manual.

activeContext.md

Purpose: Current work focus

Update Frequency: Very frequently (after every significant change)

Contains: Current focus, recent changes, next steps, decisions in progress

Your most dynamic file - reflects immediate state.

progress.md

Purpose: Status tracking

Update Frequency: Frequently (as features complete)

Contains: Completed work, in-progress items, planned features, known issues

Shows project health and what's left to build.

Best Practices

1. Start with the Brief

Fill out projectbrief.md thoroughly before other files. It's your foundation.

2. Keep activeContext.md Current

This is your most important file for day-to-day work. Update it frequently:

After implementing features
When changing focus
When making important decisions
When encountering challenges

3. Be Specific

Vague documentation is worse than no documentation:

Bad: "Fixed the bug in the auth system"

Good: "Fixed OAuth token refresh race condition in src/auth/token-manager.ts:156 by adding mutex lock"

4. Include File References

Always include file paths with line numbers:

Authentication is handled in `src/auth/oauth.ts:45-120`

This helps Claude (and humans) find the code quickly.

5. Document the Why

Don't just document what you did - document why:

Bad: "Using PostgreSQL"

Good: "Using PostgreSQL because we need ACID transactions for financial data and team has expertise"

6. Update Regularly

Don't let memory bank get stale. Update after:

Completing features
Making architectural decisions
Discovering new patterns
Changing dependencies
Shifting focus

Advanced Usage

Adding Custom Context Files

Create additional files in .claude/memory-bank/ for complex areas:

.claude/memory-bank/
├── [core files...]
├── features/
│   ├── authentication.md
│   └── payment-processing.md
├── integrations/
│   ├── stripe-api.md
│   └── sendgrid-api.md
└── testing/
    └── e2e-strategy.md

Reference them in claude.md for automatic loading.

Custom Slash Commands

Add more commands in .claude/commands/:

# Create a review command
cat > .claude/commands/review.md << 'EOF'
---
description: Review and improve code quality
---

Review the codebase for:
1. Code quality issues
2. Security vulnerabilities
3. Performance optimizations
4. Documentation gaps

Provide specific recommendations with file references.
EOF

Use with: /review

Project Intelligence Growth

The .claude/claude.md file grows smarter over time. Add sections like:

## Project-Specific Patterns

### Error Handling
- Always use custom AppError class
- Include error codes and user messages
- Log with structured logging

### API Design
- RESTful conventions
- Consistent response format: {data, error, meta}
- Version all endpoints: /api/v1/...

Claude will learn and apply these patterns automatically.

Team Usage

For team projects:

1. Onboarding: New members read memory bank to understand project 2. Knowledge Sharing: Memory bank is living documentation 3. Decision History: systemPatterns.md explains why things are built certain ways 4. Consistency: Everyone works from same understanding

Adding Your Own Skills

1. Create a new directory under skills/ (e.g., skills/my-new-skill/) 2. Add a SKILL.md with YAML frontmatter (name and description) 3. Add any supporting reference files (e.g., frameworks, archetypes, analysis guides) 4. Register the skill path in .claude-plugin/marketplace.json 5. Push to GitHub

Troubleshooting

Claude Doesn't Have Context

Problem: Claude asks about basic project info despite memory bank

Solutions: 1. Verify .claude/claude.md has alwaysApply: true 2. Check that memory bank files are filled out (not just templates) 3. Try explicitly saying "Read the memory bank first" 4. Restart Claude Code session

Memory Bank Out of Date

Problem: Documentation doesn't match current code

Solution:

update memory bank

Then work with Claude to update each file systematically.

Too Much Information

Problem: Memory bank files are overwhelming

Solution:

Focus on activeContext.md and progress.md for daily work
Keep projectbrief.md concise (1-2 pages max)
Use additional context files for detailed specs
Remember: better to have some context than none

Planning Workflow Not Working

Problem: /plan command doesn't trigger properly

Solutions: 1. Verify .claude/commands/plan.md exists 2. Check file has proper YAML frontmatter with description 3. Try restarting Claude Code 4. Use alternative: just say "Let's plan this feature carefully"

FAQ

Do I need to fill out all memory bank files?

Start with projectbrief.md and activeContext.md. Add others as your project grows. Some context is better than none.

How often should I update the memory bank?

activeContext.md: After every significant change
progress.md: When features complete or status changes
Others: When architectural decisions are made or patterns emerge

Can I use this with existing projects?

Absolutely! Just: 1. Copy .claude/ directory to your project 2. Fill out memory bank files based on current state 3. Start using Claude Code with full context

What if my project changes significantly?

Update projectbrief.md first (it's the foundation), then cascade changes to other files. Say "update memory bank" and work with Claude to revise documentation.

Is this only for Claude Code?

While designed for Claude Code, the memory bank structure works great for:

Human team members (excellent documentation)
Other AI tools
Project handoffs
Future reference

Contributing

Found a way to improve this repo? Contributions welcome!

1. Fork the repository 2. Create a feature branch 3. Make your improvements 4. Submit a pull request

Ideas for contributions:

New skills
Additional memory bank templates
More slash commands
Documentation improvements

License

MIT License. See LICENSE for details.

---

Smirnov Labs

Built and maintained by Smirnov Labs. We help teams build smarter with AI-powered workflows and developer tooling.

If this repo saved you time or sparked ideas, we'd love to hear from you -- reach out at smirnovlabs.com or open an issue to say hello.

pr-utilities

Summary

Install to Claude Code