Your AI wizard — every task you do twice becomes a spell you can cast in any tool.
Runs in Claude Code, Cursor, Codex, Gemini, OpenCode, and Copilot CLI.
25 spells · 12 framework skills · 12 workflow shapes · 6 AI tools · 0 telemetry
Maintained by RedHunt Labs
Install · Quick start · Library · Commands · Docs
Your AI assistant is good at one-off tasks. It is much better when you give it a recipe to follow.
At its core, Wizard is a meta-builder — it turns the way you already work into reusable spells your AI assistant can cast on demand. Around that core it ships three more things:
A meta-builder that interviews you and turns your recurring tasks into your own spells, saved to a personal library. (The main act.)
Capture a chat that worked —
/capture-this-chatreads the current AI session and turns it into a draft spell in one command. Or use/build-spell --from-transcript <path>for a saved transcript from any source.A spellbook of 25+ ready-to-use spells — for emails, research, decisions, planning, status updates, dev workflows, and more.
A multi-tool plugin that runs the same skills in every major AI coding tool — install once, use anywhere.
| Problem | What this fixes |
|---|---|
| You re-explain the same task to the AI every time | Spells are reusable recipes the AI loads automatically |
| Prompts work in one tool, break in another | One skill format runs in 6 different AI tools |
| The AI is great at code but useless at your work | Spells cover research, writing, decisions, daily tasks |
| You want the AI to follow a discipline (no shipping without tests, no citing without verifying) | Discipline-kind spells enforce non-negotiable rules with hard gates |
| You hand-write the same workflow as a system prompt over and over | The meta-builder interviews you once and ships a tested spell |
| If you are... | You probably want... |
|---|---|
| A researcher / writer / analyst | The research/ track — literature-scan, interview-synthesis, verifying-before-citing, and the general-research-loop chain |
| A knowledge worker | The work/ track — preparing-for-a-meeting, writing-a-status-update, responding-to-feedback |
| An everyday user | The everyday/ track — writing-an-email, summarizing-a-document, planning-a-trip, troubleshooting-whats-not-working |
| A developer | The dev/ track — brainstorming-a-feature, writing-an-implementation-plan, executing-a-plan-step-by-step, plus the dev-tdd-loop chain |
| Building a custom workflow | The meta-builder. Type /build-spell and answer 5-15 minutes of questions. |
Tailored 5-minute intros: for-non-devs.md · for-devs.md
Pick the AI tool you already use. Each install is one or two commands.
Quickest path (Claude Code or Copilot CLI):
/plugin marketplace add redhuntlabs/wizard /plugin install wizard@wizardOther tools below.
Claude Code
Recommended (one command, no clone needed):
/plugin marketplace add redhuntlabs/wizard
/plugin install wizard@wizard
Or install from a local clone:
git clone https://github.com/redhuntlabs/wizard.git/plugin install <path-to-cloned-repo>
The boot skill loads automatically via hooks/hooks.json on every session start.
Cursor
git clone https://github.com/redhuntlabs/wizard.gitIn Cursor: Settings → Plugins → Install from path → select the cloned folder.
The descriptor lives at .cursor-plugin/plugin.json.
Codex CLI
git clone https://github.com/redhuntlabs/wizard.git ~/wizard
cd ~/wizard && npm installAdd to ~/.codex/AGENTS.md:
@~/wizard/AGENTS.mdFull instructions: .codex/INSTALL.md
Gemini CLI
gemini extensions install <path-to-cloned-repo>Tool-name mapping: GEMINI.md. Descriptor: gemini-extension.json.
OpenCode
In opencode.json:
{
"plugins": ["<path-to-cloned-repo>/.opencode/plugins/wizard.js"]
}Full instructions: .opencode/INSTALL.md
GitHub Copilot CLI
Recommended (one command, no clone needed):
copilot plugin marketplace add redhuntlabs/wizard
copilot plugin install wizard@wizardFallback (older Copilot CLI without plugin support): clone the repo and add to your project's or home directory's AGENTS.md:
@<path-to-cloned-repo>/AGENTS.mdThen ensure scripts are runnable: cd <path-to-cloned-repo> && npm install. If slash commands aren't supported in your version, ask in plain English: "build a spell for me."
1. Cast a bundled spell.
Cast the writing-an-email spell to draft a reply to my landlord
saying I'll be 3 days late on rent because of a paycheck delay.
2. Build your own.
/build-spell
The meta-builder interviews you (5-15 minutes), drafts a SKILL.md, runs a try-it test, and saves it to ~/.wizard/.
3. Browse what shipped.
/list-spells
Filter by --kind discipline, --audience researcher, or --updates.
|
For anyone, any day.
|
Knowledge-worker essentials.
|
For making harder calls.
|
|
Citation-rigorous workflows.
|
For shipping software.
|
Multi-stage, end-to-end.
|
25 hand-crafted spells + 12 framework skills + 12 workflow shapes — all in spells/ and skills/.
| Command | What it does |
|---|---|
/build-spell |
Interview-driven meta-builder. Routes to the right specialist by kind. Now supports --from-transcript <path>. |
/capture-this-chat |
Turn the current chat session into a reusable spell. See docs/capturing-chats.md. |
/cast-spell <name> |
Cast a specific spell on the current task |
/list-spells |
Browse the spellbook. Filters: --kind, --audience, --updates |
/refine-spell <name> |
Update an existing spell (mandatory re-test) |
/share-spell <name> |
Export. Modes: --strict (vanilla SKILL.md), --bundle (zip) |
Start here: for-non-devs.md or for-devs.md
| Topic | Doc |
|---|---|
| Why behavioral engineering, the 1% rule, the Iron Law | philosophy.md |
kind, complexity, audience — picking the right shape |
taxonomy.md |
| The SKILL.md format spec | spell-format-spec.md |
| Building blocks: excuses tables, hard gates, warning signs | skills-primitives.md |
$WIZARD_HOME — your personal library |
personal-library.md |
| Semver, update discovery, customization preservation | versioning-and-updates.md |
| Three export modes | sharing-spells.md |
| End-to-end walkthroughs (research + dev) | general-parity-tutorial.md |
| V1 release criteria + the self-tests | definition-of-done.md |
Project structure
wizard/
README.md this file
AGENTS.md core agent instructions (Codex / OpenCode / Copilot)
CLAUDE.md, GEMINI.md tool-specific entry points
CONTRIBUTING.md, LICENSE
CHANGELOG.md release notes
package.json Node project + CLI scripts
gemini-extension.json
.claude-plugin/ Claude Code plugin descriptor
.cursor-plugin/ Cursor plugin descriptor
.codex/INSTALL.md
.opencode/ OpenCode plugin (wizard.js)
hooks/ session-start bootstrap
scripts/ validate-spell / new-spell / build-loader / adversarial tests
templates/ SKILL.md templates per kind
agents/ subagent definitions (e.g. spell-tester)
commands/ slash command definitions
skills/ framework skills (meta-builder, validator, inference engine)
spells/ bundled spellbook — everyday / work / thinking / research / dev / chains
docs/ all documentation
assets/ banner + demo SVGs
PRs welcome — see CONTRIBUTING.md. The bar for the bundled spellbook is high; for narrower or specialist spells, prefer publishing a community pack (sharing-spells.md).
Every PR runs the validator and the adversarial tests. New spells must include a worked example and a PASS tester verdict.
No telemetry. Ever. Everything runs locally inside your AI tool. Your spells live in ~/.wizard/ on your machine.
Inspired by obra/superpowers by Jesse Vincent — the SKILL.md format and the Iron Law come from there.
Built and maintained by RedHunt Labs — a CTEM and attack-surface-management team that ships open-source tooling for the security community.
MIT — see LICENSE.
Made for everyone who'd rather build a tool once than re-prompt forever.