Claude Skills for GitHub

Claude skills for GitHub are ready-made procedures that absorb the routine around a repository: they write commit messages in your team's format, assemble a pull request description from the diff, run a first-pass review, and work out why a build failed. The clearest payoff is in pull request descriptions. Developers rarely spend time on them — the body ends up as "fixes" or a bare ticket link, and the reviewer has to reconstruct intent from the diff. A skill builds the description from the actual changes: what moved, why, and what to check first. Review gets measurably faster without demanding extra discipline from the author. The second most common job is a failed build: the skill reads the log, separates the cause from its consequences and names the exact step where things broke. Both jobs share one property: the skill reads the repository instead of asking the developer to retype what is already there.

The second scenario is a failed build. The skill reads the CI log, isolates the root cause out of a cascade of errors, and proposes a fix instead of handing a person a thousand lines of raw output.

The collection below covers commits, pull requests, code review and CI/CD work.

Skills in this collection

769 skills in this collection; the 24 highest-rated are listed below.

• Упаковка MCP-сервера в бандл .mcpb — build-mcpb is a Claude Code skill that guides developers through packaging a local MCP server together with its Node or Python runtime into a single installable .mcpb file. The bundle is a zip archive containing a `manifest.json`, server code, and vendored dependencies — end users install one file with no need for Node, Python, or any toolchain on their machine. The skill covers manifest structure (schema v0.4), launch configuration via `mcp_config`, install-time user settings in `user_config` with native UI controls like directory pickers, and the build pipeline using `npx @anthropic-ai/mcpb pack` for both Node and Python targets. It applies when the server must access the local filesystem, drive desktop apps, or call OS-level APIs — cases where a remote HTTP server is not an option. Security is explicitly the developer's responsibility: MCPB provides no sandbox, so path validation and privilege scoping must be handled in server code.
• Оптимизация Bazel-сборок в монорепозитории — bazel-build-optimization is a Claude Code skill that helps optimize Bazel builds for large-scale monorepos, covering remote caching and remote execution setup, custom Bazel rule authoring, and build issue debugging. It explains core Bazel concepts — targets, packages, labels, rules, and aspects — and provides production patterns for a standard workspace layout with WORKSPACE.bazel, .bazelrc, and BUILD.bazel files. Concrete templates and worked examples are stored in references/details.md. The skill is aimed at teams configuring Bazel for enterprise codebases, migrating from other build systems, or looking to speed up CI/CD pipelines through shared build artifact caching and reproducible builds.
• Оптимизация затрат в облаке AWS, Azure, GCP — cost-optimization is a Claude Code skill that reduces cloud spending across AWS, Azure, GCP, and OCI through resource rightsizing, tagging strategies, reserved instances, and cost analysis. It covers the full cost management lifecycle — from setting up budget alerts and cost dashboards to selecting the right pricing models: Reserved Instances, Savings Plans, and Spot/Preemptible instances with discounts up to 90%. The skill includes ready-to-use Terraform examples for S3 lifecycle policies, AWS Budgets, and Auto Scaling, plus architecture patterns such as serverless-first, multi-tier storage, and environment-appropriate database sizing. Built for engineers and DevOps teams looking to audit infrastructure costs, cut cloud bills, or roll out cost governance policies across a multi-cloud environment.
• Terraform — библиотека модулей для мультиоблачной инфраструктуры — terraform-module-library is a Claude Code skill that guides the creation of reusable Terraform modules for AWS, Azure, GCP, and OCI infrastructure following infrastructure-as-code best practices. Each module follows a consistent layout: main.tf, variables.tf, outputs.tf, versions.tf, an examples/ directory, and a tests/ directory with Terratest files written in Go. Built-in patterns cover VPC, EKS, RDS, S3 on AWS; VNet, AKS, and Storage on Azure; GKE and Cloud SQL on GCP; and VCN, OKE, and Object Storage on OCI. The skill is designed for platform and DevOps teams who need to standardize multi-cloud provisioning, enforce input validation with Terraform validation blocks, apply consistent resource tagging, and compose modules — for example, wiring an RDS module to outputs from a VPC module.
• Шаблоны production-пайплайнов GitHub Actions — github-actions-templates is a Claude Code skill that generates production-ready GitHub Actions workflows for automated testing, building, and deploying applications. It covers four core patterns: matrix builds across Node.js 18/20 and Python 3.9–3.12 on ubuntu, macos, and windows runners; Docker image builds with push to ghcr.io using GitHub Actions Cache for faster pipelines; Kubernetes deployments via AWS EKS with rollout verification; and filesystem vulnerability scanning with Trivy. The skill also supports reusable workflows via workflow_call, proper permissions scoping, secrets management, and approval gates for production environments. It is aimed at developers and DevOps engineers who need to set up a reliable CI/CD pipeline on GitHub Actions while following security best practices and keeping build times low.
• Шаблоны пайплайнов GitLab CI — gitlab-ci-patterns is a Claude Code skill that helps build GitLab CI/CD pipelines with multi-stage workflows, dependency caching, and distributed runners for scalable automation. It provides ready-to-use patterns for Docker image builds via docker-in-docker with pushes to GitLab Registry, Kubernetes deployments using kubectl across staging and production environments, Terraform pipelines covering validate, plan, and apply stages, and security scanning with GitLab SAST, Dependency-Scanning, and Trivy. The skill also covers per-branch and per-job caching strategies, dynamic child pipelines, and manual approval gates before production. It is aimed at teams automating testing, building, and deployment on GitLab, adopting GitOps practices, or looking to optimize the performance of existing CI/CD pipelines.
• Блокировка обхода git-хуков в Claude Code — block-no-verify-hook is a Claude Code skill that configures a PreToolUse hook to intercept and block --no-verify and --no-gpg-sign flags in git commands before they execute. The hook is defined in .claude/settings.json with a Bash matcher: it inspects the $TOOL_INPUT environment variable using a grep regex and exits with code 2 when a bypass flag is detected inside a git command, causing Claude Code to reject the tool call entirely. This prevents AI agents from skipping pre-commit hooks that enforce linting, formatting, test runs, and security scanning, as well as GPG commit-signing policies. The configuration can be applied per-project or globally via ~/.claude/settings.json, and the regex pattern is straightforward to extend with additional flags such as --force. It is intended for teams that need reliable quality gates when using Claude Code or similar AI coding agents.
• Напоминание о пересборке TypeScript в локальном форке — local-build-reminder is a Claude Code skill that automatically prompts developers to run `npm run build` after editing TypeScript files when operating from a local OMC fork. In local mode — identifiable by the `L` suffix in the `[OMC#X.Y.ZL]` HUD indicator — Claude Code serves compiled JavaScript from `dist/` rather than TypeScript source from `src/`, so any `.ts` edits are silently ignored until a rebuild occurs. The skill triggers in four scenarios: after a `src/**/*.ts` edit, when the developer wonders why a change has no effect, before restarting Claude Code with unbuilt modifications, or when an OMC command is expected to reflect new TypeScript behavior. It stays silent for `.mjs`, `.cjs`, `.md`, and `.json` files (loaded directly from disk), for sessions not running OMC locally, and whenever `tsc --watch` or `npm run dev:full` is already handling incremental rebuilds in the background.
• Параллельная разработка фич в мультиагентной команде — parallel-feature-development is a Claude Code skill that coordinates parallel feature development using file ownership strategies, conflict avoidance rules, and integration patterns for multi-agent implementation. It covers three ownership models—by directory, by module, and by architectural layer—and enforces a cardinal rule of one owner per file, with clear guidance on extracting interface contracts in TypeScript so agents can build against each other's APIs before they are ready. Three integration patterns are provided: vertical slices, horizontal layers, and a hybrid approach, along with single-branch and multi-branch workflows suited to teams of different sizes. The skill also includes troubleshooting scenarios for common blockers such as shared-file conflicts, mid-stream decomposition changes, and early-finishing agents waiting on dependencies.
• Проектирование мультиоблачной архитектуры — multi-cloud-architecture is a Claude Code skill that provides a decision framework for designing and operating architectures across AWS, Azure, GCP, and OCI. It includes service comparison tables covering compute, storage, and database categories, four architectural patterns — Single Provider with DR, Best-of-Breed, Geographic Distribution, and Cloud-Agnostic Abstraction — plus a four-phase migration strategy from assessment to optimization. To reduce vendor lock-in, the skill recommends a portable stack built on Kubernetes, PostgreSQL, S3-compatible storage, Apache Kafka, Redis, Prometheus/Grafana, and Terraform as the infrastructure abstraction layer. It is aimed at engineers designing multi-cloud systems, selecting best-of-breed services, managing cross-provider costs, or planning cloud migrations.
• Установка и настройка oh-my-claudecode — omc-setup is a Claude Code skill that installs, refreshes, or repairs oh-my-claudecode through a single canonical setup flow. On invocation it parses flags: `--local` and `--global` run Phase 1 only, targeting the project-level `.claude/CLAUDE.md` or the global `~/.claude/CLAUDE.md` respectively; `--force` skips the existing-configuration check and reruns the full wizard from scratch. When setup has already been completed — detected via `.omc-config.json` — the skill prompts the user to choose between updating CLAUDE.md only, rerunning the full wizard, or cancelling. All configuration mutations go exclusively through `scripts/setup-claude-md.sh`, which resolves the plugin cache, verifies the compiled-source handshake, and fails closed on any coordinator protocol or exit-status disagreement. Intended for developers setting up OMC via a marketplace plugin, npm global install, or a locally checked-out repository.
• Паттерны архитектуры бэкенда: Clean, Hexagonal, DDD — architecture-patterns is a Claude Code skill that implements proven backend architecture patterns — Clean Architecture, Hexagonal Architecture, and Domain-Driven Design — to produce layered codebases with clear dependency rules, interface definitions, and test boundaries. It covers the full range of DDD tactical patterns (aggregates, value objects, repositories, domain events), Clean Architecture's inward-only dependency rule, and the ports-and-adapters model that lets you swap PostgreSQL for DynamoDB without touching the domain core. Worked examples include in-memory repository adapters that make every use-case test runnable as a plain unit test with no database or Docker, plus guidance on resolving circular imports caused by use cases importing concrete adapters instead of abstract ports. Use this skill when designing a new microservice from scratch, untangling business logic from ORM models or HTTP concerns in a monolith, or establishing bounded contexts before splitting a system into independent services.
• Проектирование пайплайна развёртывания — deployment-pipeline-design is a Claude Code skill that designs multi-stage CI/CD pipelines with approval gates, security scanning, and deployment orchestration across multiple environments. It produces pipeline stage definitions with job dependencies and caching strategy, annotated rollout configurations (canary weights, blue-green switchover, rolling parameters), deep readiness probe setups, post-deployment smoke test scripts, and automated rollback triggers keyed to metric degradation. Supported deployment targets include Kubernetes, ECS, VMs, serverless, and PaaS; compatible monitoring sources cover Prometheus, Datadog, and CloudWatch. The skill's reference documentation addresses common failure modes such as a stalled Argo Rollouts canary caused by an inconclusive AnalysisTemplate, a production approval gate waiting indefinitely due to missing reviewer assignment, and database schema mismatches after a service rollback. Teams migrating platforms, implementing progressive delivery, or aiming to reduce mean time to recovery will find this skill most useful.
• Гибридные облачные сети: VPN и выделенные каналы — hybrid-cloud-networking is a Claude Code skill that configures secure, high-performance connectivity between on-premises infrastructure and cloud platforms using VPN and dedicated connections. It covers AWS (Site-to-Site VPN up to 1.25 Gbps per tunnel, Direct Connect up to 100 Gbps), Azure (VPN Gateway, ExpressRoute up to 100 Gbps), GCP (HA VPN with 99.99% SLA, Cloud Interconnect), and OCI (IPSec VPN with redundant tunnels, FastConnect). The skill includes Terraform examples for each provider, BGP routing configuration, hub-and-spoke and multi-cloud hybrid network diagrams, and CLI troubleshooting commands for AWS, Azure, and OCI. It is aimed at teams building hybrid cloud architectures, extending data centers to the cloud, meeting compliance requirements, or implementing active-active setups with automatic BGP failover.
• Диагностика и починка oh-my-claudecode — omc-doctor is a Claude Code skill that diagnoses and fixes oh-my-claudecode installation issues by running six sequential checks. It inspects the installed plugin version against the latest on npm, scans both profile-level and project-level settings.json for legacy hooks that cause duplicate execution, checks for obsolete bash hook scripts, verifies that CLAUDE.md contains OMC markers and that its version aligns with the plugin cache, confirms Ruby is available for Ralph workflows, and detects stale plugin cache entries. All ~/.claude paths respect the CLAUDE_CONFIG_DIR environment variable. The skill is aimed at developers troubleshooting oh-my-claudecode setups — especially after updates that leave version drift, missing configuration markers, or leftover legacy hooks behind.
• Автоматический релиз через GitHub Actions — release is a Claude Code skill that orchestrates a two-step release process for Vellum assistant via GitHub Actions. Step one triggers `create-release-branch.yml`, computes the version from a bump type (`patch`, `minor`, `major`, or `hotfix`), cuts a fresh `release/v<X.Y.Z>` branch from `main` HEAD, and waits for the staging `Release` run to succeed (~15–20 minutes). Step two dispatches `release.yml` on that branch to run the full production release: Git tag, GitHub Release, macOS DMG signing and notarization, npm package publishing, Docker Hub images, iOS TestFlight, platform dependency bump, and merge-back to `main`. The default bump type is `patch`; `hotfix` cuts from the latest release tag instead of `main`. This skill is built for teams that need a structured, verifiable CI/CD release flow with a mandatory staging bake before promotion to production.
• Перезапуск сервисов Vellum с учётом ветки Git — update is a Claude Code skill that restarts Vellum assistant and gateway services with branch-aware git handling. Without arguments on main it pulls from origin/main; on a feature branch it skips git ops and restarts in place; passing a branch name checks it out and pulls if a remote exists; --pull force-pulls on whatever branch is active. The workflow runs seven steps: a preflight snapshot via vellum ps, graceful shutdown with vellum sleep plus fallback process-killing on ports 7830 and 7821, the appropriate git operation, bun install for both assistant and gateway packages (only when git ops ran), vellum wake to start services, and a final health check. To prevent silent data loss, switching branches is blocked when uncommitted changes are detected.
• Аудит безопасности GitHub Actions — cicd-security is a Claude Code skill that hunts vulnerabilities across CI/CD pipelines — covering GitHub Actions workflow injection via untrusted context variables, secret exfiltration, self-hosted runner poisoning, dependency confusion, OIDC token theft, and supply chain attacks. The skill ships with cicd_scanner.sh, which scans a single repo or an entire org (up to 30 repos in parallel with --recursive depth control), and integrates sisakulint to flag unpinned actions, dangerous patterns like eval and curl|bash, and pull_request_target misuse where base-repo secrets are exposed to attacker-controlled code. Manual analysis checklists cover injectable contexts (pull request titles, issue bodies, branch names, workflow_dispatch inputs), GITHUB_TOKEN permission abuse, and self-hosted runner exposure on public repos. Aimed at penetration testers and bug bounty researchers targeting GitHub Actions, CircleCI, Jenkins, or GitLab CI environments, it chains individual CI/CD weaknesses into critical findings such as RCE on build servers or full org-secret compromise.
• Pathfinder — аудит и картирование архитектуры кода — pathfinder is a Claude Code skill that maps a codebase into feature-grouped flowcharts, identifies duplicated concerns across features, and proposes the simplest unified architecture. Acting as an orchestrator, it deploys subagents for file-tree traversal, flow tracing, and Mermaid diagram generation — each node labeled with exact `file:line` references — while keeping synthesis and architectural decisions with the orchestrator itself. Four sequential phases produce a `PATHFINDER-<YYYY-MM-DD>/` directory containing five artifacts: a feature inventory, per-feature flowcharts, a cross-feature duplication report, a unified architecture proposal, and copy-pasteable `/make-plan` handoff prompts per unified system. It is designed for teams auditing architecture before a major refactor, consolidating parallel queue or storage implementations, or tracing how duplicated scaffolding diverged across features.
• Автоматизация семантического версионирования и релиза — version-bump is a Claude Code skill that automates semantic versioning and the full release workflow for Claude Code plugins, updating the version string across all eight manifest files simultaneously: package.json, plugin/package.json, .claude-plugin/marketplace.json, .claude-plugin/plugin.json, plugin/.claude-plugin/plugin.json, .codex-plugin/plugin.json, plugin/.codex-plugin/plugin.json, and openclaw/openclaw.plugin.json. It runs npm run build-and-sync to regenerate artifacts and keep the local marketplace in sync, then creates a git tag, pushes it, and opens a GitHub release with generated notes. CHANGELOG.md is regenerated via a Node script that pulls release data from the GitHub API. NPM publishing is deliberately handed off to the human maintainer, who controls 2FA and credentials, so the skill stops and prompts them to run npm publish before continuing. A Discord notification is sent from ~/Scripts/claude-mem/ at the end, and the workflow only closes once git status confirms a clean working tree.
• Cloudflare MCP — серверы для управления Cloudflare — Cloudflare MCP is a set of remote MCP servers that let Claude, Cursor and other MCP clients work with Cloudflare services in natural language. The client can read account configuration, process data, make suggestions and even apply changes. Reach for it when you want to dig into application logs and analytics from a chat, diagnose a DNS problem, check Logpush job health quickly, or build a Workers application with storage, AI and compute without leaving your client. The repository contains domain-specific servers: documentation, Workers Bindings, Workers Builds, Observability, Container (a development sandbox), Browser Run (pages to Markdown and screenshots), Logpush, AI Gateway, AutoRAG, DNS Analytics, Digital Experience Monitoring, Cloudflare One CASB, Radar and the Cloudflare Blog. The recommended Code Mode server (mcp.cloudflare.com) lives in the separate cloudflare/mcp repository and gives broad API access through code execution. Every server is a remote Streamable HTTP endpoint such as https://<name>.mcp.cloudflare.com/mcp, with sign-in via OAuth. It suits people working on Cloudflare; it is not for other providers.
• GitNexus — граф знаний о коде для агентов — GitNexus is an MCP server and skill set for Claude Code that indexes a codebase into a knowledge graph of dependencies, call chains, clusters and execution flows. The agent gets a complete answer in one call, for example which functions break if you change this one, so it edits blind less often. It offers 19 tools: hybrid search, full symbol context, change impact analysis, shortest path between symbols, git diff analysis before a commit, coordinated multi-file rename, raw Cypher queries and an API route map. Installed skills cover code exploration, debugging, impact analysis, refactoring and review. Run npx gitnexus analyze in the repo root, then npx gitnexus setup to connect Claude Code, Cursor and others. You need Node.js; everything runs locally. A web UI is available for browsing the graph. The license is PolyForm Noncommercial, so commercial use needs a separate license.
• Apple container — Linux-контейнеры на Mac в Claude Code — container is a plugin for Claude Code that teaches the model the command surface of Apple's container tool and how it maps to Docker, Lima, Colima and Podman on macOS. The container tool itself creates and runs Linux containers as lightweight virtual machines on a Mac; it is written in Swift and optimized for Apple silicon. It works with OCI-compatible images, so you can pull images from standard registries and push your own. The plugin helps when you need to run a container on a Mac, translate familiar Docker commands, or understand how the tools differ. Requirements per the README: a Mac with Apple silicon and macOS 26; older versions are not supported. Install the tool by downloading the signed installer package from the releases page, then run container system start. A test run: container run --rm alpine echo hello. For macOS developers who work with containers.
• Compound Engineering — цикл разработки со знаниями — Compound Engineering is a plugin of 36 skills for Claude Code that turns coding work into a loop: brainstorm requirements, plan, build, simplify, review, then write down what you learned so the next task is easier. It includes /ce-brainstorm for shaping an idea, /ce-plan for an implementation plan, /ce-work to execute it, /ce-code-review for multi-agent review before merging, /ce-compound to save lessons into docs/solutions, /ce-debug for bug investigation, plus skills for commits, pull requests, git worktrees and browser testing. The /lfg command runs the whole pipeline hands-off. Install with /plugin marketplace add EveryInc/compound-engineering-plugin, then /plugin install compound-engineering. After installing, run /ce-setup in your project. It also works in Cursor, Codex and several other hosts. MIT license. It suits people who want the agent to stop repeating old mistakes.

FAQ

Can Claude Code open pull requests on GitHub by itself?

Yes, when the environment has GitHub access configured — through the official CLI or through the GitHub MCP server. The skill is not responsible for access but for content: what goes into the title and body, and what ends up on the review checklist.

Why use a skill for commit messages when Conventional Commits exist?

Conventional Commits define the format, not the substance. A skill records what your team considers worth explaining in a commit body: which changes need a stated reason, how to reference issues, and where one commit should have been two.

All Claude Code skills