Warning
Rapid release cycle — skills change faster than this README. Where they disagree, trust the
skills (skills/*/SKILL.md) and ROADMAP.md.
A plugin that makes Claude's work scalable, maintainable, and easy-to-understand. It even responds to you in an understandable language!
Inspiration: caveman and superpowers in a single package.
- Caveman — our own flavour of concise speaking and documentation.
- Superpowers — skills and workflows to multiply productivity, but only the essentials here to avoid decision fatigue.
AI agents build code incredibly quickly, but they (yes, especially Opus 5) are overly verbose responding, documenting, and committing.
Over time, you get a repo that is cluttered, hard-to-maintain, and difficult to scale.
/theoryy-speak [ base | functional | terse ]
- Responses lead with the answer in plain sentences, no filler, no meta commentary. i.e. no more "My point is not X, but rather Y".
/theoryy-docstyle [ base | functional | terse ]
- Terse documentation. Docs say only what the reader doesn't already know.
/theoryy-commit
- Terse commits. One line stating the effect, not the steps taken. Additional sentences only when the purpose of the commit isn't clear.
/theoryy-philosophy
- A minimal design philosophy. No new feature while the current work is under ~90% polished (daily-drivable); prefer deleting to abstracting; ship the smallest diff that works.
/brainstorm → /plan → /council,
or /initiate to let Claude decide whether to run each:
- Plan before building. Complex work gets a written plan that you can review, modify, and save for later. Major work gets a panel of councils to debate and refine that plan.
/orchestrator or /player-coach
- Delegation over grinding. Run your main session in a more powerful model (Fable/Opus), let it do the high-level structural thinking. Cheaper, faster worker models do the code-writing, the powerful manager model reviews changes.
- The two skills differ in how much the manager model does by hand
orchestrator— virtually none.player-coach— hand-edits complex tasks, delegates menial ones.
- Lead author — aier9500. Ex-business-finance; economics major with a business minor at NYU. Codes as a passion, with a Linux and IT background; understands code at the structural level.
- Assistant dev — lunear01. CS + Math at the University of Toronto; knows code at the lower level.
- Special thanks to the team members unfamiliar with code and software engineering whose feedback shaped the house style.
Note
Text below is generated via /theoryy-docstyle, see it for yourself.
- Session start. The plugin reads
ROADMAP.md(the to-do list and change log) and the local READMEs, then picks or writes a plan. - Small task: just do it.
- Big task:
/initiateruns three steps — brainstorm ideas, write a plan, then four critics argue over the plan (is the scope right, is it too complex, what could go wrong, is there a better way). You get the final say: go or no-go. - Doing the work. A manager hands each step to a helper sized to the job — a cheap one for routine edits, a solid one for real work, the strongest one for anything high-stakes. When a job splits into pieces that can run side by side, one manager can hand each piece to its own manager, who runs its own helpers.
- The whole way through. One style keeps chat replies short, one keeps written docs short, one keeps commit messages short, and one philosophy decides whether a new feature is even worth adding.
- Afterwards. The to-do list and the READMEs get updated so the next session starts caught up.
- The repo is the plugin:
skills/holds the 11 slash commands,agents/holds the 10 helpers they can call,hooks/loads the workflow at the start of a session and keeps the chat style active,core/workflow.mdis the file that gets loaded in. - A manager can only hand work down two levels deep — a helper can't call more helpers of its own.
- Install:
claude plugin marketplace add Theory-Y/theoryy-workflow, then enabletheoryy-workflow.
Not released yet. Licensed under AGPL-3.0.