A Claude Code plugin that makes engineering work compound
EveryInc’s Compounding Engineering Plugin brings a methodical, AI-assisted workflow to feature planning, implementation, and review. The project—hosted on GitHub with 1.6k stars and 134 forks—aims to ensure that each engineering task leaves the codebase better prepared for the next task rather than adding to maintenance burden.
The core idea
Compounding engineering centers on a simple principle: each unit of work should make subsequent units easier. Instead of accumulating technical debt, the workflow encourages documenting patterns, producing reusable components, and codifying conventions so that knowledge compounds across the team.
How it works: three focused stages
The plugin exposes a three-step workflow through Claude Code commands, each designed to reduce friction and raise long-term quality.
-
Plan —
/compounding-engineering:plan
Turns feature descriptions into structured GitHub issues. The command researches the repository for similar patterns, consults framework documentation, produces acceptance criteria, and generates code examples that align with existing conventions. -
Work —
/compounding-engineering:work
Executes plans using isolated git worktrees, breaks work into trackable todos, and performs continuous validation. Tasks are executed systematically with tests and quality checks after each change, and a pull request is created when the plan completes. -
Review —
/compounding-engineering:review
Performs exhaustive, multi-agent reviews by checking out the PR in an isolated worktree and running 12+ specialized review agents in parallel. Reviews surface security, performance, architectural, and language-specific concerns and convert findings into trackable todos.
Additional commands include /compounding-engineering:triage, /compounding-engineering:resolve_todo_parallel, and /compounding-engineering:generate_command for triaging findings, resolving multiple todos concurrently, and creating new Claude Code commands from descriptions.
Agents and automated expertise
The plugin bundles 17 specialized agents spanning code review, security, performance, architecture, research, and workflow automation. Highlights include:
- Language-focused reviewers such as kieran-rails-reviewer, kieran-typescript-reviewer, and kieran-python-reviewer.
- Quality and safety agents like security-sentinel, performance-oracle, and data-integrity-guardian.
- Research and pattern analysis agents such as repo-research-analyst, framework-docs-researcher, and git-history-analyzer.
- Workflow helpers including every-style-editor and pr-comment-resolver.
Together, these agents aim to make planning more informed, work more repeatable, and reviews more thorough.
Installation and getting started
Installation can be performed via the Claude plugin marketplace or with a single CLI command:
-
Add the marketplace:
/plugin marketplace add [https://github.com/EveryInc/every-marketplace](https://github.com/EveryInc/every-marketplace) then installcompounding-engineering`. -
One-command install via the Claude Plugins CLI:
npx claude-plugins install @EveryInc/every-marketplace/compounding-engineering
Factory (Droid) users can convert marketplace commands to Droid format with bunx droid-factory, enable sub-agents in settings, and use the converted source without adding it to Claude Code.
Why it matters for development workflows
The plugin emphasizes repeatable processes over ad-hoc fixes. By creating issues with built-in research, using isolated worktrees for safe development, and running parallelized, specialist reviews, the workflow aims to capture learnings and surface issues earlier. Over repeated cycles, that documentation and automation intends to reduce decision fatigue and accelerate future work.
Learn more about the philosophy in action at Every’s write-up: https://every.to/source-code/my-ai-had-already-fixed-the-code-before-i-saw-it
Original source: https://github.com/EveryInc/compounding-engineering-plugin
