Donkey Developer — Claude Plugins

Overview

A marketplace of plugins for Claude Code.

The marketplace catalogue (.claude-plugin/marketplace.json) registers each plugin available in this repository. Each plugin lives under plugins/<name>/ with its own manifest, skills, agents, and prompts.

Marketplace Structure

.claude-plugin/
└── marketplace.json                ← Catalogue of available plugins

plugins/<name>/
├── .claude-plugin/plugin.json      ← Plugin manifest (name, version, description)
├── skills/{domain}/SKILL.md        ← Orchestrators (user-triggered slash commands)
├── agents/{domain}-{pillar}.md     ← Self-contained subagents (compiled, not hand-edited)
├── prompts/                        ← Source of truth (human-editable)
│   ├── shared/                     ← Cross-domain content
│   └── {domain}/                   ← Domain-specific content
└── scripts/
    └── compile.sh                  ← Builds agents and skills from prompt sources

Agents are compiled — scripts/compile.sh concatenates shared and domain prompts into self-contained agent files. A pre-commit hook runs compile.sh --check to ensure compiled output stays in sync with prompt sources.

Plugins

donkey-review

Domain-specific code reviews across Architecture, Security, SRE, and Data Engineering.

Features:

• Comprehensive code review across four domains with 16 specialist subagents
• Critical hygiene warnings
• Progressive level reporting, allowing you to focus on the next-best-action
• Summary roll-up reporting for leadership oversight of system health and maturity

Slash commands:

• /donkey-review:all — run all four domain reviews and produce a combined report
• /donkey-review:sre — SRE review (response, observability, availability, delivery)
• /donkey-review:security — Security review (authn/authz, data protection, input validation, audit & resilience)
• /donkey-review:architecture — Architecture review (code, service, system, landscape)
• /donkey-review:data — Data Engineering review (architecture, engineering, quality, governance)

Installation

Prerequisites: Claude Code must be installed.

Add the marketplace:

/plugin marketplace add https://github.com/donkey-developer/claude-plugins.git

Install the donkey-review plugin:

/plugin install donkey-review@donkey-developer

Usage

Run a comprehensive code review:

/donkey-review:all

This dispatches all 16 subagents in parallel, then synthesises the results into a combined report.

Run a single domain review:

/donkey-review:sre

Output reports are written to .donkey-review/<batch-name>/ in your working directory.

Configuration

TODO — configuration options will be documented as the plugin matures.

Development

Planning

Start an interactive planning session. The agent follows PLAN.prompt.md using Socratic elicitation to generate a plan folder with a master BRIEF.md and specific task files. A GitHub Milestone for the planned work with each task file having a target GitHub Issue. No code is written.

./scripts/plan.sh
./scripts/plan.sh "Build the code review plugin"

Or interactively in Claude Code:

Follow PLAN.prompt.md

Execution

Run a single task (interactive, one at a time):

Follow EXECUTE.prompt.md for plan/{milestone}/tasks/01-scaffolding.tasks.md

Run all tasks for a single issue (automated Ralph Wiggum loop):

./scripts/execute-issue.sh plan/{milestone}/tasks/01-scaffolding.tasks.md

Run all issues for a milestone (automated, sequential):

./scripts/execute-milestone.sh plan/{milestone}/tasks

For parallel issues, run their execute-issue.sh loops concurrently instead of using the sequential milestone script. The last task file (*-close.tasks.md) handles milestone completion — updating specs, deleting the plan directory, and closing the GitHub Milestone.

Testing

TODO

Contributing

Guidelines for contributing to this project.

License

[Choose a license]

Authors

• Lee Campbell LeeCampbell.com
• Claude Code with Opus 4.6

Support

How to get help or report issues.