On this page

Recommended flow

Five steps from a fresh install to a reviewed, shared change: install once, set up each project once, then repeat steps 3 and 4 for every bigger task.

This page is for anyone new to brewcode who wants one path instead of 27 skills. Every step names the exact command and links the page that explains it. Skip what you already have: /brewcode:setup-status shows what is installed.

The flow at a glance

Select a step to jump to it. Steps 3 and 4 repeat: review each phase, then return to step 3 for the next one.

Step 1: Install once

Why: the plugins hold the setup, publish and update skills used in steps 2 and 5; the mods add the ++ tokens and the plan-mode effort that steps 3 and 4 rely on.

Run these in a terminal, then restart Claude Code (exit, then claude). /reload-plugins alone is enough only when claude update changed nothing.

claude update
claude plugin marketplace add https://github.com/kochetkov-ma/claude-brewcode
claude plugin install brewmod-placeholders@claude-brewcode
claude plugin install brewmod-plan-effort@claude-brewcode
claude plugin install brewcode@claude-brewcode
claude plugin install brewdoc@claude-brewcode
claude plugin install brewtools@claude-brewcode
claude plugin install brewui@claude-brewcode

The mods also need node on the PATH. To let Claude run the commands for you, paste the prompt from Installation: it checks the Claude Code version, then installs both mods and the four plugins. More on each mod: Prompt placeholders, Plan-mode effort.

Step 2: Set up each project once

Why: each setup writes something into the project (a code index, conventions, agents, a review skill), so every later prompt starts from your codebase instead of generic defaults. Run them in this order, each in its own fresh session, and read the generated files before you commit them.

Order	Command	What it adds
1	/brewcode:semble-setup install	Semantic code search, so agents find code by meaning
2	/brewcode:convention-setup install	Your conventions and etalon classes in .claude/convention/, loaded through a rule
3	/brewcode:teams-setup install core "one agent per bounded context"	Domain agents built on those conventions
4	/brewcode:superreview-setup install	A project /superreview skill, used in step 4
5	/brewdoc:docsync-setup install	Docs-heavy repos only: tracks which documents went stale

Check progress at any point with /brewcode:setup-status. It lists what is installed, stale or missing, prints the exact command for each, and runs no setup itself. It also offers to enable Claude Code’s task tools, which ++m uses for its task graph in step 3. The remaining setups (task board, agent guards, memory sync) are in Full Setup.

Step 3: Start every bigger task with ++m

Why: ++m turns the main session into a manager. It builds a task graph and hands the work to agents, so the main context stays free for orchestration.

Type the token anywhere in the prompt box, followed by the task:

++m Add CSV export to the invoice list; follow the existing export pattern

Claude receives your prompt unchanged plus a hidden instruction block. To read exactly what is sent, put the caret on the token or run /placeholders show ++m.

• To get a design first: switch to plan mode (Shift+Tab twice from the default mode, or /plan), then send ++m <task>. The plan opens with the manager role and the full task graph. The plan-effort mod raises plan-mode requests to xhigh by default and never changes your session effort.
• For a long run: add +++: ++m +++ <task>. Claude creates one session cron for the current task, hourly by default, that re-reads the goal and reports drift. In plan mode it plans the cron as a step instead.

Tokens combine, except that ++rr wins over ++r. ++a (architecture first) and your own tokens work the same way. The Prompt placeholders page opens with a recorded terminal session of ++ in action; Plan-mode effort has one for plan mode.

Step 4: Review as you go

Why: a review after each phase finds regressions before the next phase builds on them.

• After a significant change: ++r check the diff. A multi-agent review, then every finding is reproduced before anything is fixed.
• After each phase: ++rr review phase 1. A review on three axes: regression first, project standard, functional correctness. At task end it asks for a final cross-review of all phases.
• Before merge: /superreview "deep review of the CSV export". The skill generated in step 2 runs your build, lint and test gates and an intent check (was the delivered change the asked one?) by default. A depth word such as deep adds domain experts, scope checks and a validator that re-checks every finding. See superreview-setup.

Step 5: Share and stay current

Why: a link lets a colleague read the result without your repo, and an update keeps the skills and mods current.

Share a file, folder or text as a link:

/brewdoc:publish docs/csv-export.md

/brewdoc:publish uploads to brewpage.app and returns a URL. Publishing is unlisted by default: anyone who has the link can open it.

Stay current with /brewtools:plugin-update: check only reports versions, a bare call walks through the update. Run /reload-plugins afterwards (restart Claude Code instead after claude update). Plugin updates do not replace files a setup already copied into your project, so run /brewcode:setup-status after an update and follow its upgrade commands. A paste-in update prompt for Claude Code is in UPDATE.md on GitHub.

One task, start to finish

The same flow on one feature, using only what steps 1 and 2 installed.

Related

Update: /brewtools:plugin-update - FAQ