EveryInc/compound-engineering-plugin

What is Compound Engineering?

Compound Engineering is an AI‑agent plugin that ships 36 ready‑to‑use skills for code‑centric AI assistants (Claude Code, Cursor, Codex, Grok, etc.). The plugin structures a development cycle into a repeatable loop – brainstorm → plan → work → simplify → review → compound – and automatically writes the lessons learned from each change back into the repository so the next iteration starts with that knowledge already baked in.

---

Core idea – the compound loop

Step	Skill	What it does
Brainstorm	/ce‑brainstorm	Interactive Q&A that extracts pure requirements and creates a unified “requirements‑only” plan.
Plan	/ce‑plan	Turns the brainstorm output into a concrete implementation plan ready for execution.
Work	/ce‑work	Executes the plan, either directly or by delegating to a cross‑model author, while handling commits and verification.
Simplify	/ce‑simplify‑code	Refactors the freshly written code for clarity and reuse before any review.
Review	/ce‑code‑review (and /ce‑doc‑review)	Multi‑agent, report‑only review that checks the implementation against the plan.
Compound	/ce‑compound	Persists the insights, constraints, and patterns discovered during the cycle into docs/solutions/ (or a configurable location).

Running the loop once writes a learning artifact; running it again reads that artifact as grounding, so the agent “remembers” past fixes, edge‑cases, or architectural decisions.

---

How you use it

1. Install the plugin in your preferred AI‑coding host (Claude Code, Cursor, Codex, Grok, etc.). The README provides exact commands for each platform – most are a one‑liner like /plugin marketplace add EveryInc/compound-engineering-plugin or codex plugin add compound‑engineering@compound‑engineering‑plugin.
2. Run the setup command in any repo:

   /ce‑setup
   

   This creates a .compound‑engineering/config.yaml, adds example artifacts, and updates .gitignore.
3. Execute the loop manually:

   /ce‑brainstorm <feature description>
   /ce‑plan
   /ce‑work
   /ce‑simplify‑code
   /ce‑code‑review
   /ce‑compound
   

4. Or hand the whole pipeline to the agent with the autonomous skill lfg ("let‑f‑go") after a brainstorm:

   /ce‑brainstorm describe the feature
   /lfg
   

   lfg runs the entire loop, pushes a PR if a remote exists, watches CI, and stops before merging unless you explicitly approve.

---

Skill catalog (high‑level groups)

Group	Representative skills	Purpose
Core loop	ce‑brainstorm, ce‑plan, ce‑work, ce‑simplify‑code, ce‑code‑review, ce‑compound	The six steps that compose every iteration.
Around the loop	ce‑strategy, ce‑product‑pulse, ce‑sweep, ce‑compound‑refresh	Keep the loop grounded in higher‑level goals and existing knowledge packs.
On‑demand	ce‑ideate, ce‑bakeoff, ce‑pov, ce‑debug, ce‑explain, ce‑doc‑review, ce‑optimize, ce‑prototype	Quick utilities for specific needs (debugging, explaining, prototyping, etc.).
Git workflow	ce‑commit, ce‑commit‑push‑pr, ce‑babysit‑pr, ce‑resolve‑pr‑feedback, ce‑worktree	Automate committing, PR creation, and shepherding feedback.
Autonomous	lfg	Run the full pipeline hands‑off.
Testing & design	ce‑test‑browser, ce‑test‑xcode, ce‑polish, ce‑dogfood	Verify and polish the output.
Collaboration	ce‑proof, ce‑handoff, ce‑promote	Share work and transition ownership.
Utilities	ce‑setup, ce‑noslop, wtf, ce‑retune, ce‑riffrec‑feedback‑analysis	Miscellaneous helpers for setup, writing, and maintenance.

---

Philosophy

• Make each unit of work easier for the next one – knowledge is captured instead of buried in commit history.
• 80 % planning & review, 20 % execution – heavy upfront thinking reduces technical debt and repetitive re‑learning.
• Compound Packs (experimental) – you can define organization‑wide rule sets that are automatically consulted during planning and enforced during review, giving you cross‑repo consistency.

---

Installation across ecosystems

The repo ships native manifests for dozens of agents (Claude Code, Cursor, Codex CLI, Kimi, Cline, Grok Build, Devin, GitHub Copilot, Factory Droid, Qwen Code, OpenCode, Pi, oh‑my‑pi, Antigravity). Installation is usually a single command that adds the repository as a marketplace source and then installs the compound‑engineering plugin. Detailed steps for each platform are in the README; the pattern is the same:

1. Register the marketplace (if required).
2. Install the plugin.
3. Restart or reload the agent so the new /skill commands appear.

---

Getting started quickly

/ce‑setup               # bootstrap config in the repo
/ce‑brainstorm make background job retries safer
/ce‑plan
/ce‑work
/ce‑simplify‑code
/ce‑code‑review
/ce‑compound

Or, fully autonomous:

/ce‑brainstorm describe the feature
/lfg

The plugin will create a PR, run tests, and stop before merging unless you approve.

---

Where to learn more

• Skill documentation – docs/guides/README.md (full catalog, chaining diagrams).
• Blog posts – “Compound engineering: how Every codes with agents” and “The story behind compounding engineering” (links in the README).
• Demo assets – assets/demo/ shows a real session replayed across two weeks, illustrating the knowledge‑compounding effect.

---

Bottom line

Compound Engineering is a real, installable plugin that turns AI coding agents into a self‑documenting development assistant. By forcing every change through a structured loop and persisting the lessons learned, it aims to reduce technical debt and make future work faster and safer.