Every Inc.'s compound-engineering plugin bets that prompts deserve the same rigor as code — compiled, tested, versioned, and shipped to ten agent platforms from one source.
The first is 39 skills and 43 agent personas that encode Every's internal loop — strategize, ideate, plan, execute, review, and the step it's named for, compound: write down what was learned so neither humans nor future agents learn it twice. The second is an ~8,900-line TypeScript CLI that ports the whole plugin to a dozen platforms from one source.
We ran the plugin's own test suite and a live multi-target conversion of the actual plugin. 1,669 of 1,678 tests passed in fourteen seconds — the nine that didn't all need live GitHub access the sandbox blocks, and not one was a real defect. Then a conversion to OpenCode, Codex, and Gemini went through with genuine semantic remapping, not passthrough.
Your team is split across Claude Code, Codex, Cursor, OpenCode, and Gemini, and every process doc rots separately for each. Compile the plugin — or your own — to all of them at once and let the lookup tables carry the translation. Installs don't clobber anything.
git clone EveryInc/compound-engineering-plugin # compile → opencode · codex · gemini · non-destructive
A real multi-target conversion of the live plugin succeeded against OpenCode, Codex, and Gemini on our bench — with genuine semantic remapping, not passthrough: a model: inherit persona converted with an inferred temperature of 0.1.
Boundary — the mappers translate, but hand-pinned aliases and targets still drift with the platforms.
You want serious agent-driven code review without adopting anyone else's whole methodology. Take ce-code-review apart and keep three bolts: reviewer selection by judgment, a confidence gate that hides anything below an explicit score, and autofix classes that draw the automation line on purpose.
ce-code-review
# 14 personas · gate at 75 · autofix vs. human-only
The orchestration is real in source and the suite backs it: 1,669 of 1,678 tests passed in 14 seconds on our bench — fourteen reviewer personas picked by reading the diff, a default gate at 75.
Boundary — the nine skipped tests need live GitHub the sandbox couldn't give them.
Your team re-solves the same class of bug every quarter because the fix lives in a closed PR thread. Switch on the Learnings loop, feed it your first ten solved problems, and come back in a quarter with one question: is review starting to cite them?
learnings/*.md # YAML frontmatter # retrieved grep-first inside future reviews
The loop ships working — the repo carries 31 of its own Learnings and retrieves them during review. Whether the compounding pays off for your team is unproven: the efficacy evidence is testimony from its authors, not measurement.
Boundary — run your own before/after; no one has measured the payoff yet, including Every.
| Play | Tier | Effort | Receipt |
|---|---|---|---|
| 01 Ship one workflow to ten platforms | Verified | AN AFTERNOON | live conversion to OpenCode/Codex/Gemini · real remapping |
| 02 Steal the review panel | Verified | A WEEKEND | 1,669/1,678 tests in 14s · 14 personas · gate at 75 |
| 03 Bank every solved problem | Demonstrated | A SPRINT | loop ships · 31 Learnings retrieved · payoff is testimony |
If you run one play tonight, run the converter — compile the plugin to every platform your team uses and let the lookup tables carry the translation. It's the first tool we've tested that hands part of the process-fragmentation tax back, down to an inferred temperature on a model: inherit persona.
git clone https://github.com/EveryInc/compound-engineering-plugin # compile to opencode · codex · gemini — backups + deep-merge