Best Practices
for Everyone
The world's knowledge is in your AI.
The world's practice is not.
Install like a package manager. Enforce like a linter.
1000+ verified best practices across 26 domains — engineering, law, medicine, finance, and more. Works with every major AI assistant.
Claude best practices, Codex best practices, Copilot best practices, OpenCode best practices, OpenClaw best practices, Gemini best practices — one library, every AI assistant.
$ curl -fsSL https://raw.githubusercontent.com/jeffreytse/grimoire/main/scripts/install.sh | bash $ grimoire update $ grimoire install
Free & open-source · MIT license · 60 seconds to install
The Problem
Your AI knows everything.
Knowing and doing are different.
McKinsey charges $1M for frameworks your AI already has access to. A lawyer charges $800/hour for contracts. A doctor charges $300 for advice rooted in evidence your AI can find. Grimoire encodes the expert's playbook — the 10,000-hour practitioner's exact steps — and makes it available to everyone.
- ✗ "Write a business strategy" → generic advice
- ✗ "Review my contract" → vague suggestions
- ✗ "Debug my architecture" → surface observations
- ✗ "Plan my workout" → generic fitness tips
- ✓ Surfaces McKinsey 7S + guides you through MECE decomposition
- ✓ Surfaces ABA clause standards + guides negotiation step by step
- ✓ Surfaces Google SRE runbooks + guides architecture review
- ✓ Surfaces NSCA periodization + guides your training program
See It In Action
One question. Four practices you didn't know applied.
Ask your AI about a real problem. Grimoire surfaces the right expert standard and guides you through applying it.
Prompt: "We're launching our SaaS in 48 hours. I'm terrified something will break. What do we do?"
Surfaced and applied — 4 practices:
How It Works
From zero to expert in 4 steps
Install once, everywhere
One command installs grimoire into every AI agent on your system. Claude Code, Copilot CLI, Cursor, Codex, Gemini CLI, OpenCode, OpenClaw, and Antigravity — all detect it automatically.
Intercepts before you ask
You don't have to know the right prompt. When you start any task, Grimoire silently checks for applicable best practices and applies or offers them proactively — before you even describe a problem.
Ask your AI naturally
You don't need to know which best practice applies. Just describe your problem — the AI finds the matching skill and routes through it automatically.
Get expert-grade output
Your AI follows the same frameworks as top institutions — McKinsey, Google, WHO, ABA, NSCA — and produces actionable, verifiable, specific output. Not summaries. Steps.
Declare your quality standard once in grimoire.toml. Best Practice Driven Development (BPDD) checks every artifact against it automatically — like a linter, for practices instead of syntax.
$ grimoire check Respects Your Choices
Grimoire learns what you've already chosen.
Already using a practice intentionally? Audit your existing work to detect what's applied. Pin preferences at the session, project, or global level — routing skills respect them and never suggest alternatives for pinned choices.
26 Domains
Every field. Every expert.
A McKinsey engagement costs $1M. A senior lawyer bills $800/hr. The practices they follow are not proprietary — they belong to everyone. Browse 1000+ skills across every professional domain, free.
Universal
Works with every AI assistant
Install once. Every AI agent on your system picks it up automatically.
Installation
Step 1 — Install the CLI
Cross-platform binary. Pick your install method.
curl -fsSL https://raw.githubusercontent.com/jeffreytse/grimoire/main/scripts/install.sh | bash
Downloads the grimoire binary and places it in /usr/local/bin. Supports macOS (amd64, arm64) and Linux (amd64, arm64).
irm https://raw.githubusercontent.com/jeffreytse/grimoire/main/scripts/install.ps1 | iex
Runs in PowerShell. Downloads and installs the grimoire binary automatically.
go install github.com/jeffreytse/grimoire@latest
Requires Go 1.21+. Installs the grimoire binary to $GOPATH/bin.
/plugin marketplace add jeffreytse/grimoire-core
Step 1: register the marketplace inside Claude Code. Then run /plugin install grimoire@grimoire-core to install all 1000+ skills. No CLI needed.
Step 2 — Install skills
Guided setup, or do it manually — your call.
grimoire wizard
Guided setup — detects your installed AI agents and walks you through registry and domain choice. No config editing required.
Or set up manually
grimoire update # clone skill library → ~/.grimoire grimoire install # install to all detected agents
Detects Claude Code, Copilot CLI, Codex, Gemini CLI, OpenCode, OpenClaw, Cursor, and Antigravity. Installs 1000+ skills to all found agents.
Quality Standard
What makes a grimoire skill
Every skill passes 4 gates before merge. No vague advice. No generic tips.
Actionable
Reader can do it immediately. Steps, not descriptions. Commands, not concepts.
Proven
Adopted by top-tier institutions — Google, McKinsey, WHO, ABA, NSCA. Cited, auditable, third-party verified.
Specific
Names tools. Cites numbers. Shows exact commands. One skill, one concept. No overlap.
Consensus
Encodes majority position among experts. No 50/50 debates. Stable, durable knowledge.
You don't need to be a developer.
Any expert with 10,000 hours of practice can contribute a skill — a doctor, a chef, a lawyer, a coach, a musician. If you know the playbook, encode it.
Plain Markdown. Follow the template. Add your 10,000-hour steps.
Automated validation + human review. Proven, actionable, specific.
Merged to main. Available to every AI agent globally. Forever.
FAQ
Common questions
What is Grimoire?
Grimoire is a free, open-source skills manager for best practices — 1000+ skills across 27 domains (engineering, law, finance, design, health, and more), each sourced from a real institution or study. Install once, and your AI routes to the right skill automatically.
Is Grimoire free?
Yes. Grimoire is free and open-source under the MIT license, published at github.com/jeffreytse/grimoire.
How is Grimoire different from a prompt library or just asking the AI directly?
Prompt libraries collect ad-hoc instructions with no verification. Every Grimoire skill must cite a real institution, pass a quality review, and prove it changes behavior — the difference between telling an AI to "act like an expert" and handing it the expert's actual, sourced playbook.
Does Grimoire enforce best practices, or is it just a skill library?
Both. Grimoire is a skills manager for best practices — install, manage, and update skills like npm — with a built-in compliance linter. Run grimoire check to verify any artifact against your declared standards, the same way ESLint checks code style. See BPDD for the full methodology.
How do I install Grimoire?
Run curl -fsSL https://raw.githubusercontent.com/jeffreytse/grimoire/main/scripts/install.sh | bash, then grimoire wizard — it detects your installed AI agents and walks you through setup. See the Getting Started guide or your platform's page above for exact commands.
What AI assistants does Grimoire support?
Claude Code, GitHub Copilot CLI, Gemini CLI, Google Antigravity, OpenCode, OpenClaw, Codex CLI, and Cursor — install once and every detected agent picks it up automatically.
How many best-practice skills are included, and where do they come from?
1000+ skills across 27 domains, each cited to a named institution, study, or professional standard — McKinsey, Google, WHO, ABA, NSCA, and more — not generic advice.
Can I contribute my own best-practice skill?
Yes. Anyone with real expertise can contribute — a doctor, lawyer, engineer, or coach. Every submission passes an automated and human quality review (source, actionability, specificity, consensus) before merging.
Does Grimoire send my data anywhere?
No. Skills run entirely inside your AI assistant, client-side — Grimoire doesn't call any external API or server. grimoire check --independent can optionally call an LLM provider for deeper AI-judged compliance checks, but only if you enable it.