Maestro Skills (maestro-skills)

A unified skills library for AI coding agents, packaged as a universal plugin for DeepSeek Harness (DSH), Claude Code, Codex CLI, OpenCode, and GitHub Copilot — two halves in one bundle:

• Domain skills — Govard development-environment orchestration and its supported web frameworks (Magento 2, Laravel, Symfony, WordPress, generic PHP): architecture, linting, performance auditing, security scanning, code review, Hyvä/Luma frontend, backend APIs, plus php-dev-core and diagram-studio.
• Process skills — the complete superpowers methodology forked verbatim from v7.0.0 (MIT, © Jesse Vincent): brainstorming, test-driven development, systematic debugging, writing/executing plans, subagent-driven development, code-review collaboration, session diagnosis, and more. See THIRD-PARTY-NOTICES.md for license and sync policy.

Every skill follows the open Agent Skills standard (a SKILL.md file with name/description frontmatter), which all major AI Agent tools understand.

Works best with Govard. Most command examples in the domain skills assume a Govard-managed dev environment (govard sh -c "...", govard tool magerun ...) — without it, you'll need to adapt them to your own container/CLI setup. One-line install: curl -fsSL https://raw.githubusercontent.com/ddtcorex/govard/master/install.sh | bash.

📂 Directory Structure

Every skill lives under skills/<name>/SKILL.md — this is the single source of truth across all supported AI tools and platforms:

maestro-skills/
├── README.md                        # Global documentation and guide
├── package.json                     # NPM package & DSH Cordis Plugin manifest (@ddtcorex/maestro-skills)
├── THIRD-PARTY-NOTICES.md           # Superpowers fork attribution + MIT license text
├── scripts/
│   └── sync-superpowers.sh          # Refresh the superpowers fork from upstream
├── tsconfig.json                    # TypeScript compiler configuration
├── src/                             # DSH Cordis Plugin source code
│   ├── index.ts                     # Cordis plugin entrypoint: serves the packaged skills (the agent preset is a declaration row, see Installation)
│   └── dsh-types.d.ts               # DSH type definitions
├── .dsh-plugin/                     # DeepSeek Harness Agent Preset definition
│   ├── preset.yml                   # Web GUI preset display metadata ("Maestro Skills")
│   └── agent.cordis.yml             # DSH Agent preset composition
├── .claude-plugin/                  # Claude Code plugin manifests
├── .codex-plugin/                   # Codex CLI plugin manifest
├── .agents/plugins/                 # Codex marketplace manifest
│
└── skills/                          # one folder per skill (`ls skills` is the source of truth)
    ├── 🧠 PROCESS SKILLS (forked from obra/superpowers v7.0.0)
    │   ├── brainstorming/              # Socratic design refinement with approval gates
    │   ├── diagnosing-superpowers/     # Diagnose what went wrong in a session
    │   ├── test-driven-development/    # RED-GREEN-REFACTOR iron law
    │   ├── systematic-debugging/       # 4-phase root-cause process
    │   ├── verification-before-completion/  # Evidence before success claims
    │   ├── writing-plans/ + executing-plans/
    │   ├── subagent-driven-development/ + dispatching-parallel-agents/
    │   ├── requesting/receiving-code-review/
    │   ├── using-git-worktrees/ + finishing-a-development-branch/
    │   ├── writing-skills/             # How to author new skills
    │   └── using-superpowers/          # Skill-system introduction
    │
    ├── 📦 CORE STANDARDS & ARCHITECTURES
    │   └── magento2-dev-core/           # Magento 2 core guidelines (DI, Repositories, Security)
    ├── 🛠️ QUALITY ASSURANCE & AUDITING
    │   ├── magento2-linter/             # PHPCS, PHPStan, PHPMD quality checks
    │   ├── magento2-performance-audit/  # Web vitals, infrastructure and DB profiling
    │   ├── magento2-security-scan/      # Static vulnerability code scanning
    │   └── magento2-code-review/        # PR/module/theme/project review orchestration
    ├── 🎨 FRONTEND & BACKEND FRAMEWORKS
    │   ├── magento2-hyva-dev/           # Alpine.js, Tailwind CSS, CSP payment pages
    │   ├── magento2-frontend-dev/       # Luma Knockout.js, LESS, RequireJS
    │   ├── magento2-backend-dev/        # REST, GraphQL resolvers, Cron, Queues
    │   └── maestro-design/              # Tailwind + shadcn/ui design systems, a11y, Figma to code
    ├── 🐘 PHP FOUNDATION
    │   └── php-dev-core/                # PSR-12, Composer, PHPStan, framework-agnostic PHP
    ├── 📐 DIAGRAMS
    │   └── diagram-studio/              # Mermaid + editorial HTML diagrams, CLI verifier
    └── 🔧 DEV ENVIRONMENT & CLI TOOLS (Govard Stack)
        ├── govard-toolbox/              # Base container orchestrator toolbox
        ├── govard-magento/              # Magento-specific dev env commands
        ├── govard-laravel/              # Laravel-specific dev env commands
        ├── govard-symfony/              # Symfony-specific dev env commands
        ├── govard-wordpress/            # WordPress-specific dev env commands
        └── review-in-worktree/          # Govard commands in detached review worktrees

🤖 Skill Compatibility Matrix

Every skill works identically on every listed tool.

💾 Installation

1. DeepSeek Harness (DSH) — as a plugin (recommended)

One command is a complete install — the plugin serves every packaged skill itself:

dsh plugin --profile web add github:ddtcorex/maestro-skills

Then restart dsh web. No install.sh needed.

The agent preset is a separate, explicit step. Skills and presets are two different
mechanisms: the plugin registers a skill provider, but a preset is an
@deepseek-ai/dsh-agent-preset declaration row that some bundle has to carry. To get
"Maestro Skills" in the Web GUI agent picker, copy the two files from this package's
.dsh-plugin/ into a dsh-agent-preset row in the profile's cordis.patch.yml:

- insert:
    - id: preset-maestro-skills
      name: '@deepseek-ai/dsh-agent-preset'
      config:
        id: maestro-skills
        name: "Maestro Skills"
        description: "AI Agent preset bundling the maestro-skills library"
        plugins:          # the entry list from .dsh-plugin/agent.cordis.yml
          - id: …

This plugin used to copy that directory into ~/.dsh/.agent-presets/maestro-skills/ on
every boot and log that the preset was installed. DSH stopped reading that directory when
presets became declaration rows (2026-09-22) — the write changed nothing and the log
misled. The template still ships in .dsh-plugin/ because declaration rows are copied
from it by hand.

Codex + Claude delegation preset (optional bundles, optional preset): when the web
profile also installs the two optional subagent bundles (@deepseek-ai/dsh-subagent-codex
and @deepseek-ai/dsh-subagent-claude-code), add a second declaration row for
"Maestro Skills + Subagents (Codex + Claude)" — the same composition with
subagent_codex and subagent_claude_code tool rows enabled, so the model can delegate to
those products through their official app-server/SDK result contract instead of whichever
CLI happens to be on PATH. src/subagent-variant.ts is the executable definition of that
derivation. Both providers authenticate via their native subscription login, so
quota-limited accounts may see provider-side quota errors until the limit resets.

The package builds itself on install via its prepare script. pnpm ≥ 10 blocks git-dependency build scripts until you allow them — if the first add reports an ignored build, copy the key pnpm prints into the profile's pnpm-workspace.yaml and re-run:

allowBuilds:
  '@ddtcorex/maestro-skills': true

To reference the plugin from a hand-written patch layer instead, the row must name the real package:

- insert:
    - id: maestro-skills
      name: '@ddtcorex/maestro-skills'

Do not link skills into ~/.dsh/skills/ (the old install.sh --target dsh
path, removed): the preset-layer skill-filesystem provider would shadow
the plugin with a stale cache. The plugin is the single source of truth
on DSH.

2. Claude Code — as a plugin

/plugin marketplace add ddtcorex/maestro-skills
/plugin install maestro-skills@ddtcorex

Updates: /plugin marketplace update ddtcorex.

3. Codex CLI — as a plugin

codex plugin marketplace add ddtcorex/maestro-skills
codex plugin add maestro-skills@ddtcorex

Updates: codex plugin marketplace upgrade ddtcorex.

4. OpenCode & GitHub Copilot — loose skill files (no plugin loader there)

curl -fsSL https://raw.githubusercontent.com/ddtcorex/maestro-skills/master/install.sh | bash

The installer only serves tools without a plugin loader. On Claude Code,
Codex CLI and DSH, install the plugin (sections 1–3 above) instead of
linking loose files: a loose copy shadows the plugin with a stale cache.

🔄 Superpowers fork maintenance

The process skills are an upstream fork, not hand-maintained copies:

scripts/sync-superpowers.sh            # latest upstream HEAD
scripts/sync-superpowers.sh v7.0.0     # a specific tag

The script prints a diff for review; the fork carries no content additions (its PRESERVE list is intentionally empty), and the fork-provenance note in using-superpowers/SKILL.md is re-applied by hand during review. Do not edit forked skill bodies by hand — see THIRD-PARTY-NOTICES.md.

⚡ Extension Guide: Adding New Domain Skills

You can easily expand this hub with any new framework or toolchain supported by Govard (e.g., Laravel, Symfony, React, Vue, Docker, etc.).

Step 1: Create the Skill Folder

Create skills/[framework/tool]-[purpose]/ using kebab-case (e.g., skills/laravel-dev-core/, skills/vue-hyva-checkout/). For process skills, don't create them here — contribute upstream to obra/superpowers instead.

Step 2: Create a SKILL.md File

Every skill folder must contain a SKILL.md file at its root with valid YAML frontmatter:

---
name: your-skill-name
description: |
  Describe exactly when the AI Agent should trigger this skill.
compatibility: claude, codex, opencode, copilot, dsh
depends: [any-dependencies-if-applicable]
---

# Your Skill Title

## Capabilities
Describe what this skill enables the agent to do.

## Best Practice Patterns
Provide code templates, standards, and typical commands.

## Verification
Step-by-step commands to verify output works properly.

One bundle, one workflow: Govard domain expertise plus superpowers process discipline. 🚀

Contributing

Changes follow the Superpowers workflow mandated in AGENTS.md:
brainstorming → writing-plans → executing-plans with strict TDD. Feature
branches only; never commit to master directly.

Credits & Attribution

• Domain skills: © DDTCoreX, MIT License (see LICENSE).
• Process skills: forked from obra/superpowers v7.0.0 by Jesse Vincent / Prime Radiant, MIT License — full notice in THIRD-PARTY-NOTICES.md. The fork carries no local content additions.