Custom Claude Code plugins and skills

Personal collection of Claude Code plugins and skills for structured, high-quality development workflows. Augments the official Claude plugins marketplace.

Installation

Add this marketplace to Claude Code:

/plugin marketplace add DmitriyYukhanov/claude-plugins

Then install plugins by name. The @ suffix is the marketplace name dmitriy-claude-plugins (from this repo's marketplace.json), not the GitHub path:

/plugin install issue-to-pr@dmitriy-claude-plugins
/plugin install codex-collaboration@dmitriy-claude-plugins

You can also run /plugin to browse and install interactively.

Plugins

Plugin details

agent-teams

Set up and orchestrate Claude Code agent teams with:

• Auto-activating skill for agent team setup and coordination
• Teams vs subagents decision guide
• Display modes (in-process and split pane)
• Prompt templates for code review, debugging, and exploration
• Configuration reference with architecture, permissions, and limitations

View documentation

codex-collaboration

Cross-model collaboration between Claude and Codex with two workflows:

• collaborative-loop — Sequential drive/validate/act cycles. Claude analyzes, Codex validates each finding, both models must agree before any action is taken. Iterates until approved or max rounds.
• cross-review — Parallel dual review. Both Claude and Codex review independently. Findings confirmed by reviewer agreement, cross-validation, or evidence research are auto-applied each round; only genuinely inconclusive disagreements surface for user decision.

The default Codex model is gpt-5.6-sol (requires Codex CLI 0.143.0 or newer); if unavailable it falls back to gpt-5.5, then gpt-5.4.

Requires the Codex plugin for Claude Code. Install it first:

1. /plugin marketplace add openai/codex-plugin-cc
2. /plugin install codex@openai-codex
3. /codex:setup to verify

View documentation

humanizer

Auto-routing EN + RU humanizer for Claude Code that removes signs of AI-generated writing:

• Auto-detects language by Cyrillic ratio (≥60 % → RU, ≤10 % → EN, otherwise asks)
• English ruleset — vendored from blader/humanizer, 29 patterns from Wikipedia AI Cleanup
• Russian ruleset — vendored from ilyautov/humanizer-ru, 44 patterns with hard bans and triple-pass audit
• Explicit overrides ("humanize as English" / "обработай как русский") and mixed-text handling
• Optional voice calibration from a writing sample

Distinct from the Claude.ai-only humanizer.skill listed below — this is a Claude Code plugin with auto-routing.

View documentation

implementation-prd

Implementation-ready spec bundle authoring with:

• Auto-activating skill for turning requests into build-ready spec bundles
• Templates for product, system, and feature PRDs, contracts, schemas, data models, and test plans
• init-spec-bundle.sh scaffolding script for standard 4-file bundles
• Quality gates checklist and cross-file alignment rules
• Worked example from a real desktop app project

View documentation

issue-to-pr

Drive a single GitHub issue from triage to a merge-ready pull request through a gated pipeline:

• One flow for a bare issue or a card on a GitHub Projects (v2) board — Step 0 resolves the input and decides whether board-status sync applies.
• Each run works in its own ../<repo>-worktrees/issue-<N> git worktree, so concurrent local agents on different issues never clash.
• Hard gates that block forward progress: design hardening (cross-review or a multi-agent fallback), tests green (typecheck + tests, plus visual checks for UI), and a code-review loop that runs until clean or five passes.
• Scales the work to the task's tier (trivial through epic): research depth, design (an autonomous design panel for complex work), review level, and report length all size to the issue — and it asks at most one batched question, deciding everything else itself and surfacing those decisions in the report and PR.
• Takes on more than a single issue: a plain request with no number is drafted into an issue and run; a from-scratch, epic-sized request is decomposed into dependency-ordered child issues, each shipped through its own gated PR; and after a merge, an optional smoke check on the base opens a draft revert PR (never an automatic rollback) if the change broke something.
• A robust merge gate: it reads any GitHub review first (a changes-requested review or an unresolved thread reroutes as a change request), and a behind-base PR is updated and re-checked automatically — but re-merged without asking again only when the base merge left the PR's own diff untouched.
• Opens the PR and stops; once you approve it in-session it squash-merges, then deletes the branch, tears down the worktree, and clears the run's temp artifacts (keeping anything important).
• The PR auto-links the issue (Closes #N) to close on merge; board cards advance to in-progress at branch cut and in-review at PR open, with Done left to GitHub's merge-time automation.
• Graceful by default — missing the project token scope degrades to link-only, and a failed status write never blocks the PR.
• The /issue-to-pr:tune skill reads the friction log that runs leave behind and turns it into batched improvements to the pipeline.
• Optional .claude/issue-to-pr.local.md for board URL, base branch, and test commands (auto-detected when unset); companion skills (superpowers:*, /deep-research, /cross-review, humanizer, /code-review) used if installed, with inline fallbacks otherwise.

View documentation

learning-guide

Generate self-contained, offline-first, interactive HTML learning guides for any artifact:

• analyze skill — reads input artifact(s) (codebase, planning session, refactor plan, generic doc), writes a tour-spec.json describing sections, embedded sources, cross-refs, quizzes, and external link maps. Hands off to render.
• render skill — runs the bundled zero-dependency Node renderer to produce index.html plus launcher scripts. Idempotent for generated artifacts; re-runnable from the shell after hand-edits to the spec.
• learning-guide entry-point skill — explains the flow and dispatches to the right step.

Requires Node.js on PATH. No npm install step.

View documentation

lsp-setup

Set up LSP (Language Server Protocol) for Claude Code projects with:

• Auto-detection of project languages from source files and project markers
• Automatic installation of language server binaries and Claude Code LSP plugins
• Environment-aware setup (MINGW, WSL, macOS, Linux) with gotcha handling
• State machine workflow that resumes from any point (install → restart → validate)
• Post-restart validation confirming LSP operations work per language
• Supports all 12 official LSP plugins (C#, Python, TypeScript/JS, Go, Rust, Java, Kotlin, Lua, PHP, Ruby, Swift, C/C++)

View documentation

python-dev

Python development workflow with:

• /python-dev command for end-to-end workflows (discovery → architecture → implementation → review → testing)
• Architecture patterns, protocols, and Mermaid diagrams
• PEP 8-aligned coding guidelines and type-hint best practices
• pytest-based testing patterns, fixtures, mocking, and coverage guidance
• Python-specific review agent that can delegate to general feature review

View documentation

tg-alerts

Add Telegram error/alert notifications to any project:

• Interactive 7-phase setup: @BotFather bot creation, chat ID discovery, code generation, integration, testing
• Reference implementations for Python async (FastAPI), Python sync (Django/Flask), and Node.js/TypeScript (Express/NestJS)
• Built-in deduplication, HTML formatting, graceful failure handling, and fire-and-forget delivery

View documentation

tg-voice

Transcribe Telegram voice messages using local Whisper:

• PostToolUse hook automatically transcribes .oga voice messages downloaded via the Telegram channel plugin
• /voice-to-text-config skill for guided setup: faster-whisper installation, model download, and end-to-end verification
• Configurable model size via WHISPER_MODEL env var (default: base)

Requires: faster-whisper (pip install faster-whisper) and the Telegram channel plugin.

View documentation

typescript-dev

TypeScript development workflow with:

• /typescript-dev command for full-stack TypeScript workflows
• Architecture design for modules, interfaces, and test stubs with Mermaid diagrams
• TypeScript coding guidelines, async patterns, and framework-aware checks
• Testing patterns for unit, integration, and E2E tests (Jest/Vitest style)
• TypeScript-specific review agent that can delegate to general feature review

View documentation

unity-dev

Unity C# development workflow with:

• /unity-dev skill orchestrating the full workflow: discovery, architecture, implementation, review, testing
• unity-run skill for CLI execution (builds, tests, method execution, asset imports) with Unity installation auto-detection and log monitoring
• /unity-tests-run skill running Unity Test Framework tests via CLI batchmode
• Architecture design with Mermaid diagrams, C# coding guidelines following Microsoft conventions, EditMode/PlayMode test patterns
• Unity-specific review, simplification, and test-runner agents

View documentation

Claude.ai Skills

Skills for the Claude.ai web interface (not Claude Code). Install via the Skills section in Claude.ai settings.

humanizer

Rewrites AI-generated text to sound more natural and human. Converted from blader/humanizer for Claude.ai usage.

File: skills/claude.ai/humanizer.skill

License

MIT