Personal collection of Claude Code plugins and skills for structured, high-quality development workflows. Augments the official Claude plugins marketplace.
Add this marketplace to Claude Code:
/plugin marketplace add DmitriyYukhanov/claude-pluginsThen install the plugins you want to use:
/plugin install agent-teams
/plugin install codex-collaboration
/plugin install humanizer
/plugin install implementation-prd
/plugin install issue-to-pr
/plugin install learning-guide
/plugin install lsp-setup
/plugin install python-dev
/plugin install tg-alerts
/plugin install tg-voice
/plugin install typescript-dev
/plugin install unity-devYou can also install individual plugins directly, for example:
/plugin install python-dev@DmitriyYukhanov/claude-plugins
/plugin install typescript-dev@DmitriyYukhanov/claude-plugins| Plugin | What it does |
|---|---|
| codex-collaboration | Cross-model Claude + Codex collaboration — sequential drive/validate loops and parallel dual review |
| implementation-prd | Turn feature requests into build-ready spec bundles with PRDs, contracts, schemas, and test plans |
| agent-teams | Orchestrate multiple Claude Code instances working in parallel with shared tasks and messaging |
| issue-to-pr | Drive a GitHub issue or Project board card through a worktree-isolated, gated pipeline to an approved, merged PR |
| tg-alerts | Add Telegram error/alert notifications to any project — guided setup with reference implementations |
| tg-voice | Auto-transcribe Telegram voice messages using local Whisper via PostToolUse hook |
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
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.
Requires the Codex plugin for Claude Code. Install it first:
/plugin marketplace add openai/codex-plugin-cc/plugin install codex@openai-codex/codex:setupto verify
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.
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.shscaffolding script for standard 4-file bundles- Quality gates checklist and cross-file alignment rules
- Worked example from a real desktop app project
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.
- 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, withDoneleft to GitHub's merge-time automation. - Graceful by default — missing the
projecttoken scope degrades to link-only, and a failed status write never blocks the PR. - Optional
.claude/issue-to-pr.local.mdfor 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.
Generate self-contained, offline-first, interactive HTML learning guides for any artifact:
analyzeskill — reads input artifact(s) (codebase, planning session, refactor plan, generic doc), writes atour-spec.jsondescribing sections, embedded sources, cross-refs, quizzes, and external link maps. Hands off to render.renderskill — runs the bundled zero-dependency Node renderer to produceindex.htmlplus launcher scripts. Idempotent for generated artifacts; re-runnable from the shell after hand-edits to the spec.learning-guideentry-point skill — explains the flow and dispatches to the right step.
Requires Node.js on PATH. No npm install step.
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++)
Python development workflow with:
/python-devcommand 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
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
Transcribe Telegram voice messages using local Whisper:
- PostToolUse hook automatically transcribes
.ogavoice messages downloaded via the Telegram channel plugin /voice-to-text-configskill for guided setup: faster-whisper installation, model download, and end-to-end verification- Configurable model size via
WHISPER_MODELenv var (default:base)
Requires: faster-whisper (pip install faster-whisper) and the Telegram channel plugin.
TypeScript development workflow with:
/typescript-devcommand 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
Unity C# development workflow with:
/unity-devcommand for full workflow orchestration- Architecture design with Mermaid diagrams
- C# coding guidelines following Microsoft conventions
- EditMode/PlayMode testing patterns
- Unity-specific code review agent
- Code simplification agent
Skills for the Claude.ai web interface (not Claude Code). Install via the Skills section in Claude.ai settings.
Rewrites AI-generated text to sound more natural and human. Converted from blader/humanizer for Claude.ai usage.
File: skills/claude.ai/humanizer.skill
MIT