Pack the AI memory.Install it into any project.
Structure your AI sessions' memory, context, plans, and operating know-how as rulepack.json + Markdown — then publish, version, and install it across every project and agent with one command.
Install any pack in one command
$pnpm dlx rulepack add nextjsWorks with every agent you use
The problem
Every agent reads a different file.
Claude Code wants CLAUDE.md. Cursor reads .cursor/rules. Codex, Jules and the rest read AGENTS.md. Maintain them by hand and they drift the moment your standards change.
- CLAUDE.mdedit each
- .cursor/rules/main.mdcedit each
- AGENTS.mdedit each
- .windsurfrulesedit each
- .clinerulesedit each
Five files. Five copies of the same rules — drifting apart.
One manifest. One command regenerates every target — versioned, reproducible, and verifiable.
Why RulePack
Everything your AI memory needs to travel.
Built like a package registry, tuned for AI memory.
One manifest, every agent
Declare your memory once in rulepack.json. The CLI writes the exact file each agent expects — no copy-paste, no drift.
Install in one command
rulepack add drops the right files into any project and records them, so the next person gets the same setup.
Immutable & versioned
Every release is semver-pinned and immutable. Upgrade deliberately — never wake up to rules that changed under you.
Open, no lock-in
An open registry of plain-text files. Fork a pack, self-host, or eject to raw files at any time.
Verified integrity
Every tarball is SHA-256 checksummed end to end, so what you install is exactly what was published.
Public, private, or paid
Ship a pack to everyone or keep it private to your scope — both live today. Mark a pack paid to gate its files; paid checkout is coming soon.
How it works
From memory to every agent in three steps.
Author
Describe your memory in one rulepack.json. rulepack init scaffolds it, plus an AGENTS.md to start from.
Publish
rulepack publish ships an immutable, versioned release to the registry under your scope.
Install
Anyone runs rulepack add and the pack's files land in their project, ready for every agent.
Author
Describe your memory in one rulepack.json. rulepack init scaffolds it, plus an AGENTS.md to start from.
Publish
rulepack publish ships an immutable, versioned release to the registry under your scope.
Install
Anyone runs rulepack add and the pack's files land in their project, ready for every agent.
One source of truth
One manifest. Every memory file.
One rulepack.json points at the real memory files in your repo — switch tabs to see each one rulepack add drops into a project.
{ "name": "@acme/nextjs", "version": "1.4.0", "license": "MIT", "visibility": "public", "memories": { "claude": "CLAUDE.md", "cursor": ".cursor/rules/main.mdc", "agents": "AGENTS.md", "custom": { "plan": "docs/plan.md" } }}# Project memory - App Router only — never the pages dir.- Server Components by default.- Validate every API boundary with zod.Explore the registry
Find a pack, or publish your own.
Popular packs
Most-installed packs across the registry
Recently published
Fresh from the registry in the last few days
Articles
Read the workflow. Install the memory.
Publish AI operating know-how and playbooks — free or paid — and link the memory pack behind each one so readers install it in a command.
Free or paid
Publish the whole piece free, or write a free preview and gate the rest — the gated body is never sent to a non-buyer. Paid checkout is coming soon.
Linked to packs
Attach the memory pack behind the article — readers go straight from the write-up to rulepack add, and the pack lists the article under Related.
Drafts & discovery
Drafts stay private to you; published articles surface in search right alongside packs.
RulePack AI
Watch RulePack AI read every file
RulePack AI reads every rule file in a published version — the manifest, README, CLAUDE.md, AGENTS.md, .cursor rules — line by line, catching prompt injection and unsafe instructions before your agent ever runs them.
- ⠋CLAUDE.md
- AGENTS.md
- .cursor/rules/nextjs.mdc
- README.md
- rulepack.json
- LICENSE
Coverage
What RulePack AI catches
Precise static rules plus a local model, tuned for the ways a rule file can attack the agent that reads it.
Prompt injection
“Ignore previous instructions”, fake system or developer messages aimed at the agent.
Credential exfiltration
Reading .env, tokens, or SSH keys and sending them to an external URL or webhook.
Unsafe commands
curl | sh, rm -rf, chmod 777 — download-and-execute or destructive shell.
Persistence
Writing shell rc files, git hooks, or crontab to run code later.
Stealth
“Don’t tell the user”, hide actions, or suppress logs.
Obfuscation
Base64/hex blobs, zero-width characters, and homoglyph tricks.
Coming soon · Pro
Earn a badge the world can trust
RulePack AI scans every published version for prompt injection, credential exfiltration, and unsafe agent instructions — then awards a verified safety badge on your pack page. A Pro feature, launching soon.
- Local, private AI — your pack is never sent to a third-party model
- Continuous re-scans as new attack patterns emerge
- A public “Safety checked” shield on your pack page
- CI gate & policy controls for teams
Issued per published version · re-scanned as new threats emerge
Pack the AI memory once. Install it anywhere.
Author once, publish to the registry, install it into any project and agent. No lock-in, every version immutable.
or start from scratch
$pnpm dlx rulepack init