Claude Code skills for better vibe coding

Open-source skills that make AI coding agents better at real work: proving a task is really done, getting unstuck from a bug loop, and catching data leakage and compliance gaps before launch. Each one came out of real projects where an agent kept making the same mistake, and one rule runs through all of them: never assume anything. They are free, MIT licensed and on GitHub.

Packs

done-gateloop-breakercompliance-checkskill-scoutdoc-extractionmake-automationhonest-model-evalhackathon-sprintandroid-device-testexam-sheet

Pick a pack to see its skills, then pack it up into one install. Or click a skill to install just that one.

all-skills: 10 skills

Install

Everything, with one command from your terminal:

claude plugin marketplace add iampreetdave-max/agent-skills && claude plugin install all-skills@iampreetdave-skills

On Windows PowerShell 5, use ; instead of &&. Restart Claude Code afterwards. Or do it from inside Claude Code, choosing a pack or a single skill:

/plugin marketplace add iampreetdave-max/agent-skills
/plugin install senior-pack@iampreetdave-skills

Want just one skill? Click it above or below: each has its own page with a one-skill install command. Install one pack or single skills, not several that overlap, or the same skill loads twice. Without the plugin system, copy a folder from skills/ in the repo into ~/.claude/skills/; any agent that reads SKILL.md can use them.

Use

Each skill loads by itself when what you ask matches it, like “I'm stuck, same error again” or “is this app compliant?”. To call one directly, type / and its name, for example /all-skills:loop-breaker. Get updates with claude plugin marketplace update iampreetdave-skills.

The skills

Stops an agent calling work finished until every ask is rebuilt from your own words, proven with file:line evidence, audited by a second agent that is not told what changed, and passed through the project's own checks.
done-gate · Use when: Before anything is reported done, or when you ask "did you do everything?"
senior · student
Gets a stuck session out of a loop: stop editing, reproduce the failure by hand, test every assumption, prove the root cause with evidence, and hand off to a fresh session when the context itself is the problem.
loop-breaker · Use when: The same error keeps coming back, or you say "still broken"
senior · student
Catches compliance gaps before a vibe-coded product ships: reads the code for personal data, payments, third parties and AI features, asks only what the code cannot answer, maps India's DPDP Act, GDPR, the EU AI Act, HIPAA, PCI DSS and more, checks no secret file or internal endpoint is reachable, and pushes a prioritised fix list. Not legal advice.
compliance-check · Use when: Before launching, or when adding payments, accounts, AI features or user data
senior
Finds skills that fit your project on GitHub, ranked by fit, then health, then stars, with their open issues read. Counts which installed skills you actually use and tells you what to remove.
skill-scout · Use when: Starting a new project, or the weekly check
senior
OCR and document extraction whose numbers you can trust: the PDF's own text first, each page read with context from the one before, and validators (running balance to the cent, checksums, totals) that decide what is correct.
doc-extraction · Use when: Extracting data from PDFs, scans, statements or invoices
senior
Make.com automations done right: when to use Make, Python or both, webhooks between them, Microsoft Teams, a human approval step, never acting twice, dry runs and error handling.
make-automation · Use when: Building or fixing a Make scenario
senior
Stops vibe-coded ML models from leaking: audit the data first, check every feature exists at prediction time, split by time, beat the market, and report calibrated results with their uncertainty.
honest-model-eval · Use when: Building or reviewing a predictive model
senior
Building to win a hackathon with an AI agent: the hard category, open source and papers first, one flawless demo path, a real deployed URL, and a sleek deck with a rehearsed three-minute demo.
hackathon-sprint · Use when: Entering a hackathon or a demo day
student
Tests your app on a real Android phone over adb: installs the build, taps real UI elements, cuts the network, kills the process, and reports pass or fail with screenshots and logs.
android-device-test · Use when: "Test it on my phone"
Turns a syllabus and past papers into one printable revision page, chapters ranked by the marks they actually carried, every topic an exam-ready answer with a hand-drawn diagram.
exam-sheet · Use when: Exam prep for any subject, with notes for GTU
student

Questions

What are Claude Code skills?

A skill is a folder with a SKILL.md file of instructions, and sometimes scripts, that Claude Code loads when your request matches its description. It teaches the agent a repeatable way of working, such as proving a task is finished before saying so.

How do I install all of them?

Run this in your terminal, then restart Claude Code: claude plugin marketplace add iampreetdave-max/agent-skills && claude plugin install all-skills@iampreetdave-skills

Can I install just one skill?

Yes. Every skill has its own page with a one-skill install command, for example /plugin install loop-breaker@iampreetdave-skills from inside Claude Code.

Do they work with other AI coding agents?

They use the open SKILL.md format. Copy a skill's folder into the skills directory of any agent that reads SKILL.md files.

How do these make vibe coding better?

They target the mistakes AI coding agents make most often: calling work done when parts are missing (done-gate), looping on the same bug (loop-breaker), leaking future data into ML models (honest-model-eval), and shipping without a compliance check (compliance-check).

Are they free?

Yes. They are open source under the MIT licence, and the code is on GitHub.