12 skills · built from real production

The skills used to build a real CRM, packaged so you can drop them into your Claude.

Twelve Claude Code skills covering the full arc: anchor the build, ship the slice, catch the bugs, maintain the app. Every pattern is production-tested. The fastest install is one prompt pasted into your Claude Code — your Claude writes every SKILL.md straight into ~/.claude/skills/.

Install via Claude (one prompt) Browse the pack → ↓ Download zip
12skills
Production-tested
4phases
Build → Ship → Maintain
2AIs
Claude + Codex
1prompt
One-shot install
The Pack

12 skills, grouped by when you'd use them.

Click any card to read the full SKILL.md — the same file Claude reads when the skill fires. Hit the download icon to save just that one.

Phase 1

Anchor the build — before you write a line of code

Stop building for nobody.
Phase 2

Build the slice — actually ship code

Three big bodies of knowledge.
Phase 3

Ship safely — catch bugs before customers do

The avengers cycle.
Phase 4

Keep the app alive — maintenance & memory

Don't let it rot.
⌥ Claude + ◯ Codex (GPT)
The avengers cycle

How OpenAI's Codex got wired into the Claude Code workflow.

Claude is sharp, but Claude reviewing Claude misses bugs because they share blind spots. The fix: a different model with different training reviewing every commit. Different priors, different misses. Codex catches the bugs Claude would have happily shipped — cursor lies, NaN propagation, lifecycle races, architectural traps. Real findings, every commit.

The billing split that matters: Codex runs against a ChatGPT Plus subscription, NOT the Anthropic API. The two are completely separate. When Anthropic's 5-hour cap hits, Codex still works. When OpenAI rate-limits, Claude still works. Two flat budgets that don't compete.

1
Install
Run /plugin install openai/codex-plugin-cc inside Claude Code. CLI lands at ~/.local/bin/codex.
2
Auth
Run codex login in terminal. Browser OAuth via ChatGPT Plus. Flat-rate quota, no per-token charge.
3
Review
Before every meaningful commit, Codex reviews the diff with a focused adversarial prompt.
4
Patch or roll back
Take the L's. Same finding 3× in a row = approach is wrong; roll back, don't patch.
# The adversarial review prompt — what actually works codex exec --skip-git-repo-check " Adversarial review of [uncommitted diff] in [repo]. Context: [one paragraph — what the change does, why]. Run `git diff` to see the change. Adversarially review: 1. [specific concern A] 2. [specific concern B] 3. Anything else I missed. Be terse, list concrete bugs only, max 200 words. If clean, say so in 1 line. "
Real bugs Codex caught in production:
  • Pagination cursor that would have silently infinite-looped on a /search endpoint — same cursor name behaves differently across endpoints of the same vendor
  • Number("$245.00") returning NaN, propagating through every $-stat downstream
  • Vercel after() lifecycle: .then() chains died when the serverless worker tore down
  • quote_complete-as-bot-state-proxy lie that fired wrong "Pause Amber" buttons
  • Cursor regression detection too narrow (single-ID check missed partial overlap)
  • An architectural risk caught at design time — would have prevented a cost-runaway incident
Install

One prompt, and your Claude installs everything.

Open a fresh Claude Code session on your Mac, paste the prompt below. Your Claude fetches all 12 SKILL.md files from this page's URL and writes them straight into ~/.claude/skills/. Done.

Paste into Claude Code 
Install the Super Skill Pack into ~/.claude/skills/.

The pack is served at __BASE_URL__. Read the manifest at
__BASE_URL__/install-manifest.json — it lists 12 skills, each
with a slug and a path to its SKILL.md.

For each skill in the manifest:
  1. mkdir -p ~/.claude/skills/<slug>
  2. Fetch __BASE_URL__/<path> (the path field from the manifest)
  3. Save the response body as ~/.claude/skills/<slug>/SKILL.md

When done, list the 12 slugs you installed and tell me to start a
fresh Claude Code session so the skills load.
↓
Prefer to install manually?

Download the zip, unzip wherever, then copy the folders. Same result, no prompt.

curl -O __BASE_URL__/skill-pack.zip unzip skill-pack.zip cp -r skills/* ~/.claude/skills/
▶
After install, fire it up

Start a fresh Claude Code session, type Q-Explanation. The pack walks you through itself. The right skill auto-fires on natural phrases.

$ claude > Q-Explanation
Recommended add-on: install Codex CLI for adversarial review on every commit. Inside Claude Code run /plugin install openai/codex-plugin-cc then codex login in your terminal. Codex bills against a flat ChatGPT Plus subscription, so it never competes with your Anthropic budget. The Codex Gate skill walks the rest of the setup.