Multi (multi)

Русская версия → · 中文 →

ask Claude, Codex and OpenCode models to do same thing, get better review/answers.

command	what it does
/multi:code-review	code review by Claude sub-agents + Codex + OpenCode, judged by Claude Code into one report
/multi:check-if-done	is the work actually finished, or only finished-looking - every claim backed by a command that was really run
/multi:ask	put one question to all of them and see the three answers side by side
/multi:adhd	brainstorm wide - 5 cognitive frames, each one handed to a different model, then scored and grouped into a shortlist

Human-written README

soo basically its what the name implies - several models instead of one. four commands:

1. /multi:code-review - code-review with Claude's sub-agents, Codex request (via headless exec) and OpenCode's headless request + ponytail lens for overengineering. just read ponytail plugin desc, its kinda cool
2. /multi:check-if-done - the "no bro is it REALLY done" one. a model that just wrote the code is the worst judge of whether it works, so this asks models that didn't write it, and refuses to call anything done without actually running a command that proves it. should fix "looks done but not really done sorry lmao" cases
3. /multi:ask - just ask everyone the same thing and see three answers - u'll get three answers, no merging, no judjign
4. /multi:adhd - the ideas one. its the adhd skill trick (weird frames like "3am on-call", "biology", "10-year-old with no idea what software is" - they push the model past the first three boring answers), but here every frame goes to a different model, so the ideas spread out by frame and by model at the same time. at the end it scores them, groups them and hands u a shortlist, so u dont get 30 unsorted ideas dumped on ur head

install

claude plugin marketplace add szarkans/multi
claude plugin install multi@szkills

or
npx skills add szarkans/multi

or
git clone https://github.com/szarkans/multi ~/.claude/skills/multi

then restart claude code and just ask claude code to setup everything why bother

Why your README written like that?

im tired of ai-slop-saas-b2b-skill readme's. plain user-readable text is better, no garbage noise, only what you actually need.
and tbh just ask claude what this plugin about. we ALL do that w/o reading actual readme.

---

Evals

I mean... works for me lol. Test it yourself lmao

License

MIT © Sergei Shatrov