Claude Code Skills

A collection of 49 reusable Claude Code skills for software engineering workflows.

Claude Code Skills

A collection of 49 reusable Claude Code skills for software engineering workflows. Drop them into any project's .claude/skills/ directory and get structured, opinionated guidance for the full engineering lifecycle.

Covers architecture, code review, debugging, testing, security, deployment, CI/CD, database design, observability, and more - everything a development team needs, packaged as ready-to-use AI agent skills.

Built to the Agent Skills open standard. Compatible with Claude Code, Claude API, Claude Agent SDK, and claude.ai.

What Are Skills?

Skills are markdown files that give Claude domain-specific expertise for your project. They work like reusable prompt templates with structured instructions - Claude loads skill metadata at startup and reads full instructions on demand, so only the relevant skill enters the context window.

You can invoke skills manually (/review, /scope-check) or let Claude use them automatically when relevant. Skills work in the CLI, IDE extensions, the API, and claude.ai.

For the full spec, see Agent Skills Overview and Best Practices.

Skills (49)

Core Dev Workflow

SkillInvokeDescription
architectautoSystem architecture - diagrams, concurrency models, design decisions
reviewautoCode review - performance, correctness, security, code quality
bugautoBug investigation - categorize, trace root cause, fix, prevent
testingautoTesting strategy - what to test, how to test, coverage targets
workflowautoGit workflow - branching, commits, PRs, merge process
build/buildBuild & deploy - multi-language commands, release checklist
perfautoPerformance audit - response time, memory, CPU, database, network
best-practicesautoCode quality - separation of concerns, error handling, thread safety
scope-check/scope-checkFeature creep guard - evaluate proposals against core goals
dev-setup/dev-setupDev environment setup - prerequisites, install, configure, verify
phase-gate/phase-gateMilestone evaluation - deliverables, gate criteria, readiness
tech-stackautoTechnology decisions - evaluation framework, dependency health
telemetryautoObservability - error tracking, metrics, logging, analytics
debug-latencyautoLatency diagnosis - full request path tracing
qa/qaQA & user testing - session guides, checklists, metrics

Engineering Practices

SkillInvokeDescription
refactorautoSafe refactoring - code smells, transformations, behavior preservation
migrateautoMigration planning - database, API, dependency, platform
security-auditautoSecurity audit - OWASP Top 10, auth, input validation, secrets
api-designautoAPI design - REST conventions, errors, pagination, versioning
api-integrationautoThird-party APIs - client design, auth, webhooks, rate limits
incidentautoIncident response - triage, mitigation, post-mortems
docs/docsDocumentation - ADRs, runbooks, API docs, review
dependency-updateautoDependency updates - audit, risk assessment, security patches
deployautoDeployment - strategies, pre/post checks, rollback
estimate/estimateWork estimation - task breakdown, complexity, risk flags
onboardautoProject onboarding - codebase overview, key files, workflows
pr-authorautoPR authoring - sizing, self-review, descriptions, stacking

Infrastructure & Ops

SkillInvokeDescription
ci-cdautoCI/CD pipelines - stages, caching, secrets, optimization
database-designautoDatabase design - schema, indexing, migrations, EXPLAIN
containerizeautoContainerization - Dockerfiles, multi-stage builds, Compose
iacautoInfrastructure as Code - Terraform, state, modules, drift
monitoringautoMonitoring & alerting - golden signals, SLOs, dashboards
loggingautoStructured logging - levels, correlation IDs, aggregation
cachingautoCaching - placement, invalidation, stampede prevention
error-handlingautoError handling - retry, circuit breakers, degradation
config-managementautoConfiguration - env vars, feature flags, secrets, validation

Architecture & Data

SkillInvokeDescription
data-modelingautoDomain modeling - entities, relationships, aggregates, state machines
schema-evolutionautoSchema evolution - expand-contract, compatibility, coordination
async-patternsautoAsync workflows - queues, pub/sub, events, eventual consistency
state-managementautoClient state - server state, URL state, derived state
data-pipelineautoData pipelines - ETL/ELT, idempotency, backfill, quality checks

Reliability & Operations

SkillInvokeDescription
rollbackautoRollback - code, database, config, feature flags
load-testingautoLoad testing - targets, realistic scenarios, bottleneck analysis
feature-flag-opsautoFeature flags - progressive rollout, kill switches, cleanup
observability-debugautoProduction debugging - metrics + logs + traces correlation
git-recoveryautoGit recovery - force-push, broken rebase, lost commits
tech-debtautoTech debt - inventory, prioritize, business case, paydown

Cross-Cutting

SkillInvokeDescription
accessibilityautoAccessibility - WCAG 2.1, keyboard nav, screen readers
i18nautoInternationalization - string extraction, plurals, RTL, locales

auto = Claude invokes automatically when relevant. /command = user invokes manually.

Quick Start

# Copy all skills into your project
git clone https://github.com/pvnarp/agent-skills.git /tmp/agent-skills
cp -r /tmp/agent-skills/skills/ .claude/skills/

# Or copy individual skills
cp -r /tmp/agent-skills/skills/review/ .claude/skills/review/

Install globally (all projects)

cp -r /tmp/agent-skills/skills/* ~/.claude/skills/

Symlink (stays up to date with pulls)

git clone https://github.com/pvnarp/agent-skills.git ~/.agent-skills
ln -s ~/.agent-skills/skills/ your-project/.claude/skills

Customization

These skills are intentionally generic. Customize for your project:

  1. Edit the skill - Add project-specific conventions, tools, or patterns
  2. Add reference files - Create a reference/ subdirectory for additional context (progressive disclosure - only loaded when needed)
  3. Adjust frontmatter - Control invocation behavior

Frontmatter Reference

---
name: skill-name                    # Lowercase, hyphens, max 64 chars
description: What it does and when  # Max 1024 chars, third person
user-invocable: true                # Show in /command menu (default: true)
disable-model-invocation: true      # Only manual invocation (default: false)
allowed-tools: Read, Grep, Glob     # Restrict tool access when active
context: fork                       # Run in isolated subagent
argument-hint: "[issue-number]"     # Hint shown during autocomplete
---

Skill Structure

my-skill/
├── SKILL.md              # Main instructions (required, keep under 500 lines)
├── reference/
│   ├── patterns.md       # Detailed reference (loaded on demand)
│   └── examples.md       # Examples (loaded on demand)
└── scripts/
    └── validate.sh       # Utility scripts (executed, not loaded into context)

Authoring Guidelines

Following the official best practices:

  • Be concise - Claude is already smart. Only add context it doesn't already have.
  • Be opinionated - Take a stance. Generic advice is useless.
  • Be structured - Checklists, tables, templates. Claude follows structure well.
  • Include output formats - Tell Claude exactly how to format its response.
  • Write descriptions in third person - "Designs..." not "I help you design..."
  • Include "Use when..." - Help Claude know when to invoke the skill.
  • Keep SKILL.md under 500 lines - Split into reference files for detailed content.
  • One level of references - Link from SKILL.md to reference files, not deeper.
  • Test with real tasks - Use the skill in a real project before sharing.

Why This Exists

I kept rebuilding the same skills across different projects. Every time I started something new, I'd write a review skill, an architecture skill, a testing skill... and they'd all end up slightly different but mostly the same.

So I pulled out the ones that kept working well across projects, cleaned them up, and put them here. These have been tested across a mix of backend APIs, frontend apps, CLI tools, and infrastructure work. Some started rough and got better after I noticed where Claude was giving me weak advice and tightened the instructions.

If a skill works for you, great. If it doesn't, open an issue or PR and I'll fix it.

Contributing

See CONTRIBUTING.md for guidelines on adding new skills or improving existing ones.

License

MIT