Claude Code plugin · v1.1.0

Stop re-explaining your stack
every AI session.

A Claude Code plugin that gives the agent persistent context about your project — skills, rules, hooks, and subagents. Built for solo founders and designer-developers on a modern indie stack.

install

/plugin marketplace add amnafarzy/ravie
/plugin install ravie@ravie
View on GitHubRead the docs

the problem

Every session starts blank.

Claude Code is fast and capable inside a single session. But it doesn't know your stack, your conventions, or your safety rules until you tell it. So you tell it. Then you tell it again the next session. And the one after that.

It's not just tedious — it compounds. The agent generates code in the wrong style, suggests tools you've already ruled out, misses the safety checks you care about. You spend time on corrections that should never have been needed.

The fix isn't a better prompt. It's persistent context: a set of files the agent reads at the start of every session so it already knows where it is, what rules apply, and which workflows to follow. That's what Ravie is.

architecture

Four layers. One coherent system.

Each layer has a distinct job. You write one file. The rest runs automatically.

Companion pluginsSuperpowers + Karpathy Guidelines

HOW the agent approaches any work — brainstorm before build, TDD, surgical changes, subagent dispatch.

Workflow disciplineBehavioral baseline
Rules + Hooksrules/ + scripts/ via hooks/hooks.json

Reference guidance the agent reads when relevant, plus deterministic enforcement that Claude cannot bypass.

5 rule reference files4 hook scripts
CLAUDE.mdPer-project entry point

WHAT the agent knows about this project. Stack, commands, systems, active priorities. You write this once per project.

Your stack + commandsConnected systemsActive priorities
Skills + Agentsskills/ + agents/

33 specific workflows selected by Claude Code from YAML descriptions, plus 4 subagents in isolated context.

33 active skills4 subagents

Layers 1 and 2 are always active. Layers 3 and 4 load per project.

what's inside

Everything in one plugin install.

Skills are selected automatically from their YAML descriptions — you don't invoke them manually. Hooks enforce rules the agent cannot bypass.

33active skills

Selected by Claude Code from YAML descriptions. No manual invocation needed.

Sample skills

  • issue-to-pr
  • debug-root-cause
  • requirements-griller
  • figma-lovable-handoff
  • supabase-guardian
5rule files

Domain reference guides the agent reads when working in that area.

Rule files

  • git.md
  • supabase.md
  • ui.md
  • context-hygiene.md
  • karpathy-guidelines.md
4hook scripts

Deterministic enforcement. Claude cannot choose to ignore these.

Scripts

  • block-env-writes.sh
  • block-bash-secrets.sh
  • block-main-push.sh
  • auto-format.sh
4subagents

Spawn in isolated context to avoid confirmation bias.

Agents

  • code-reviewer
  • security-auditor
  • research-scout
  • ux-checker

vs superpowers

How is this different from Superpowers?

If you know Superpowers, the short answer: they're complementary. Superpowers owns process. Ravie owns context. Install both.

dimension
Ravie
Superpowers
Handles
Project context
Workflow discipline
Scope
Stack-specific (Supabase, Vercel, Linear, Notion)
Process-agnostic
Design skills
Yes — Figma, responsive, a11y, motion, Three.js
Not the focus
Growth / launch
Yes — positioning, SEO, CRO, launch checklist
Not the focus
System-of-record rules
Yes — Notion / Linear / GitHub governance
Not the focus
Permission tiers
Yes — 5-tier model with deterministic hooks
Not the focus

Install both. They don't overlap — Superpowers owns process, Ravie owns context.

quick start

Up and running in four steps.

Steps 1–3 get you a working install. Step 4 is recommended but optional.

1

Install the plugin

Run both commands inside Claude Code.

Claude Code
/plugin marketplace add amnafarzy/ravie
/plugin install ravie@ravie
2

Add an entry file to your project

Drop a CLAUDE.md into your project root. Fill in your stack, commands, and connected systems. Takes about 30 minutes the first time.

Terminal
cd /your/project
curl -O https://raw.githubusercontent.com/amnafarzy/ravie/main/quickstart/CLAUDE.md
# Edit CLAUDE.md — fill in your stack and commands
3

Reload and verify

Reload plugins, then ask Claude to read back what it knows about your project.

Claude Code
/reload-plugins

# Then in a new session:
"Read CLAUDE.md and tell me what you understand about this project."
4

Install companion plugins (recommended)

Superpowers handles workflow discipline. Karpathy Guidelines enforce surgical engineering behavior. Ravie handles context. All three together cover the full loop.

Claude Code
/plugin marketplace add obra/superpowers-marketplace
/plugin install superpowers@superpowers-marketplace

/plugin marketplace add forrestchang/andrej-karpathy-skills
/plugin install andrej-karpathy-skills@karpathy-skills

verify

Open a new Claude Code session in your project and ask:

"Read CLAUDE.md and tell me what you understand about this project's stack and conventions."

If it summarizes your stack correctly, it's working.

who it's for

Built for a specific kind of builder.

Solo founders

Context-switching between product, code, and ops. Ravie's daily-brief and weekly review skills help you stay oriented across sessions.

Designer-developers

Moving between Figma, code, and deployment. The figma-lovable-handoff, design-system-ui, and a11y skills cover the gap most tools ignore.

Indie hackers

Multiple projects in parallel on a shared stack. Write CLAUDE.md once per project, share skills across all of them.

Anyone tired of blank-slate sessions

If you've re-explained your stack to Claude more than twice this week, this is for you.

Star on GitHubContributing guide