EveryInc /
compound-engineering-plugin
36 coding-agent skills that plan, build, review and record what you learned
- 25.7k stars
- 301 stars since 3 Oct
- TypeScript
- Updated today
- Some experience
- Agent skills & plugins
AI coding agents tend to forget. Each new task starts from scratch, and lessons from earlier bugs and decisions are lost. Compound Engineering is a plugin that gives the agent a repeatable workflow and a place to write down what it learned.
The core loop has six steps: brainstorm requirements, plan the implementation, do the work, simplify the code, review it, then compound the learning. The last step writes a note into docs/solutions/. Later brainstorm and plan runs read those notes as background, so they don't repeat old mistakes.
The README says to put most effort into planning and review, and less into execution. Beyond the loop, the plugin has skills for debugging, git and PR handling, browser testing, and a hands-off mode called lfg. It supports 14 agent hosts, including Claude Code, Cursor, Codex, Cline and GitHub Copilot.
What it does
Six-step engineering loop
Brainstorm, plan, work, simplify, review and compound are separate skills. Each hands its output to the next, using one plan file that grows from requirements into an implementation plan.
Captured learnings
ce-compound writes what you learned into docs/solutions/. Later planning runs find those notes and carry their constraints forward.
Multi-agent code review
ce-code-review runs several review agents against the plan and reports findings. Applying fixes locally is a separate, explicit step.
Hands-off mode
After a brainstorm, lfg runs the pipeline on its own. It plans, works, reviews, applies fixes, runs browser tests and commits. If a git remote exists it opens a PR and watches CI. It won't merge unless you allow it.
Many agent hosts
Install paths are given for Claude Code, Cursor, Codex app and CLI, Kimi, Cline, Grok Build CLI and others.
What you could build
- Turn a rough feature idea into a requirements doc, a plan and reviewed code with a fixed sequence of commands.
- Record the lesson from a tricky bug so future planning runs avoid it.
- Run a multi-agent review of a change against its plan before merging.
- Hand a described feature to the lfg skill and get back a commit or a pull request.
- Share team conventions across repos by writing them as Compound Packs (experimental).
Get started
These commands are for Claude Code. Cursor and Codex have their own install steps in the README. After installing, run /ce-setup in your project, then try /ce-brainstorm with a small feature idea.
/plugin marketplace add EveryInc/compound-engineering-plugin/plugin install compound-engineeringCommands are copied from the project's README. Read it on GitHub for the full setup.
More in Agent skills & plugins
superpowers
A set of skills that makes coding agents plan, test and review before they write code
mattpocock/skills
Small, editable agent skills for planning, testing and reviewing code with AI
ECC
Agents, skills, hooks and memory that add an engineering workflow to coding agents
andrej-karpathy-skills
A CLAUDE.md file that tells Claude Code to ask first, keep code small, edit narrowly
anthropics/skills
Anthropic's example Agent Skills: folders of instructions Claude loads per task
ponytail
A skill that makes AI coding agents write the least code that works
gstack
Slash-command skills that give Claude Code roles like reviewer, QA and release engineer
addyosmani/agent-skills
A pack of 25 Markdown skills that guide AI coding agents from spec to ship
taste-skill
Agent skills that steer AI coding tools toward better layout, type and motion
archify
Agent skill that turns a description or a codebase into an interactive HTML diagram