Coder
Pack № 02 · Build
Vibe Coder
Guardrails for your first app: ask first, plan, debug properly, prove it works.
npx skills add obra/superpowers@22f0b463e7a5e579904dd6211a520321035526f7 --skill brainstorming writing-plans systematic-debugging verification-before-completion -g -y && \ npx skills add DietrichGebert/ponytail@96608e078b2d5ea607c5c3b8b5a54d1c1fba6846 --skill ponytail -g -y && \ npx skills add mattpocock/skills@abeeaf95660ba9e8b7c753eb38f441ca707a8823 --skill git-guardrails-claude-code -g -y && \ npx skills add anthropics/skills@e4bb5d960d87126ff3b3a8af1a104b7db67560ba --skill webapp-testing -g -y && \ npx skills add tt-a1i/archify@8509a3d63e763b4d454ea9aafa1f939545fcb1ea --skill archify -g -y && \ npx skills add latent-spaces/brag@06379432a9243c951dfcc051153be18d2d261044 --skill brag-slim -g -y
npx skills add obra/superpowers@22f0b463e7a5e579904dd6211a520321035526f7 --skill brainstorming writing-plans systematic-debugging verification-before-completion -g -y npx skills add DietrichGebert/ponytail@96608e078b2d5ea607c5c3b8b5a54d1c1fba6846 --skill ponytail -g -y npx skills add mattpocock/skills@abeeaf95660ba9e8b7c753eb38f441ca707a8823 --skill git-guardrails-claude-code -g -y npx skills add anthropics/skills@e4bb5d960d87126ff3b3a8af1a104b7db67560ba --skill webapp-testing -g -y npx skills add tt-a1i/archify@8509a3d63e763b4d454ea9aafa1f939545fcb1ea --skill archify -g -y npx skills add latent-spaces/brag@06379432a9243c951dfcc051153be18d2d261044 --skill brag-slim -g -y
Fetches 9 skills from 6 author repos, each pinned to the version we reviewed. Nothing is mirrored. Uses the open-source skills CLI; needs Node.js 22.20 or newer.
Opens Claude Code with the install prompt already typed. Nothing runs until you press Enter.
Skills installed from a terminal do not reach claude.ai. A one-step route for claude.ai is in the works.
Until then you can add skills one at a time: open a card below, follow the link to the author's repo, and upload that skill folder as a ZIP under Customize, then Skills.
What's inside
9 skills. Each card links to its author's repository at the exact version we reviewed.
- Brainstorming 1/9 Design before code Ready Stops the agent from building on a guess: it asks what you want and why, lays out two or three approaches, writes a short design doc and waits for your approval. @obra MIT Source obra/superpowers at 22f0b46
Brainstorming
The question you did not think to answer is the one that sinks a first app. This skill starts on its own, so you do not have to know when to call it.
- Writing plans 2/9 Planning Ready Turns an approved spec into a step-by-step implementation plan with exact files, interfaces and tests, saves it in your repo and asks you to review it before any code. @obra MIT Source obra/superpowers at 22f0b46
Writing plans
A plan you can read and approve is your chance to catch a wrong turn before it costs an afternoon of generated code.
- Ponytail 3/9 Coding discipline Ready Lazy senior developer mode: the agent solves the whole task with the least new code, reuses what the codebase has and ends each reply with what it skipped or did not check. @DietrichGebert MIT Source DietrichGebert/ponytail at 96608e0
Ponytail
Agents overbuild, and you cannot review what you do not understand. Ponytail keeps every change small and tells you what it left out.
- Systematic debugging 4/9 Debugging Ready Makes the agent find the root cause before it changes anything: read the error, reproduce it, trace the data, test one hypothesis at a time and rethink after three failed fixes. @obra MIT Source obra/superpowers at 22f0b46
Systematic debugging
The beginner loop is fix, break, fix again. This ends the guessing after three failed attempts and sends the agent back to the cause.
- Verification before completion 5/9 Verification Ready Forbids the agent to say done, fixed or passing until it has run the command that proves it in the same turn and read the output. @obra MIT Source obra/superpowers at 22f0b46
Verification before completion
A confident sentence is not a working app. With this installed you get the test output or the build result before you hear the word done.
- Git guardrails for Claude Code 6/9 Git safety Needs app Sets up a Claude Code hook that blocks destructive git commands, such as reset --hard, clean -f, branch -D and every git push, before the agent can run them. @mattpocock MIT Source mattpocock/skills at abeeaf9
Git guardrails for Claude Code
One wrong reset deletes work you do not yet know how to get back. A hook that says no is the cheapest insurance in this pack.
- Web app testing 7/9 Browser testing Needs app Has the agent write Python Playwright scripts that open your local app in headless Chromium, click through it, take screenshots and read the browser console. @anthropics Apache-2.0 Source anthropics/skills at e4bb5d9
Web app testing
Lets the agent open your app and look at it, so it stops asking you whether the button works.
- Archify 8/9 Diagrams Needs app Turns a system description or a real codebase into a checked, standalone HTML diagram you can click through: architecture, workflow, sequence, data flow, ERD and more. @tt-a1i Apache-2.0 Source tt-a1i/archify at 8509a3d
Archify
After a week of prompting you own code you have never read. Ask for a diagram of it and click through how the parts connect.
- Brag (slim) 9/9 Launch video Needs app Turns your project or a website into a 15 to 25 second launch video with music, built from your real UI and copy, plus a poster frame and share text. @latent-spaces MIT Source latent-spaces/brag at 0637943
Brag (slim)
You built something, so show it. One request turns your real screens into a short video you can post the same evening.
Good to know
What you bring
- Node.js for the optional visual companion
- Claude Code (the skill installs one of its hooks)
- bash and jq for the hook script
- Python 3
- Playwright for Python with Chromium
- Node.js 18 or newer
- Chrome, Chromium or Edge for the browser check
- ffmpeg (the skill asks before it fetches a copy)
- Chrome or another headless browser
Pick one, not both
- Requirements: brainstorming and the grilling skill by Matt Pocock, with its /grill-me command, do the same job. This pack uses brainstorming because it starts on its own, while grilling waits until you call it. Do not install both.
- Superpowers: the four Superpowers skills here are single installs. In Claude Code you can install the whole Superpowers plugin instead, which adds the plan executors and a session hook. Use one route, never both, or every skill shows up twice.
- Browser checks: webapp-testing needs Python. If you only have Node.js, use playwright-cli by Microsoft or agent-browser by Vercel in its place, never two browser skills together.
- Launch video: brag-slim builds the video with tools already on your machine. The full brag skill from the same repository renders through HyperFrames and needs its skills installed too. Pick one.
Do not combine with
These collections steer the same part of your agent's work and get in each other's way.
- Every skill file read in full
- Pinned to a reviewed version
- Author and licence on every card
- Requirements and costs disclosed
- No sponsored skills inside
Nov 2026Reviewed