Skip to content

Latest commit

 

History

48 Commits

Folders and files

NameName
Last commit message
Last commit date
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Practical AI Agent Skills That Finish the Job

GitHub stars skills.sh License: MIT

25 installable agent skills for turning ambiguous work into concrete, checkable results: plan a product, recover a confused project, stress-test AI instructions, preserve your writing voice, close an iteration, or make a playlist that actually imports.

Built from real solo-builder workflows for Codex, Claude Code, Cursor, and other tools that support the open Agent Skills format.

Start here

Choose one concrete result to try:

Your current problem First skill to try See the result before installing
A task's files moved and the next action is buried in an old chat TaskDock Run the move-and-recover demonstration
An agent workflow needs explicit budgets and failure checks Agent Loop Builder Reproduce a rejected boolean budget
A product feature is missing retries, ownership or a complete user path Product System Mapper Read a complete import-and-retry example

The first two demonstrations execute real code with fictional inputs. The product map is an authored example. None is a customer testimonial or a time-saving claim.

See every available skill:

npx skills add m1nga/skills --list

Start with one skill for the task you have now:

npx skills add m1nga/taskdock

Then ask your agent: Use $taskdock to organize this task and record its next action. Choose another skill from the task tables below when you need it. You do not need all 25 skills loaded for an ordinary task.

If a skill helps a real task, star that skill's own repository or share a small reproduction of what failed. You can also star this collection to find the full catalog again.

Pick the result you need

Define, test, and build products

You need to… Use You get
turn a vague idea into a build-ready plan Build Plan Interviewer a locked scope, decisions, and non-goals
find what will break before building Product Idea Stress Test simulated first contact, ranked risks, and a stronger concept
clarify who a product is for and why it matters Product Definition Interview a five-part definition and visible validation debt
map an entire product system Product System Mapper journeys, capabilities, ownership, gaps, and delivery slices
experience a product as a new user First-Time User Tester a severity-ranked report of confusion and fixes
decide whether to repair or rebuild Project Rebuild Advisor an evidence-based recover, repair, rebuild, continue, or stop verdict

Run reliable agent work

You need to… Use You get
check what to reuse before creating another skill Skill Reuse Checker an extend, combine, or build-new verdict
turn repeated work into a closed loop Agent Loop Builder persistent state, recovery, and independent verification
keep a complex task organized and ready to resume TaskDock a portable Desktop workspace with identity, plan, progress, and file checks
capture side work without losing the main thread Side Task Assistant bounded background work or an honest queued status
see what a coding session really finished Coding Session Recap completed, unverified, open, and next separated clearly
close a milestone cleanly Iteration Cleanup & Handoff preserved decisions, safe cleanup, and a tested takeover
collect an AI session for human review AI Work Review Pack one verified folder containing files and decisions
retain reusable lessons without stale decisions Project Lessons Log a project ledger plus portable lessons
publish a finished skill as a product Skill Publisher a searchable repository, direct install path, and release proof

Improve prompts, writing, and evaluation

You need to… Use You get
clean up dictated or rambling input Prompt Cleaner a clear reusable prompt that preserves your intent
build a marketing prompt from rough material Marketing Prompt Builder brand-, audience-, and channel-aware AI instructions
make AI drafts sound like you Writing Voice Matcher a measurable voice fingerprint and draft checks
explain something in one memorable sentence One-Sentence Explainer one clear line a stranger can repeat
think through an ambiguous decision Decision Thinking Partner a better frame, real alternatives, and evidence-based convergence
find how AI instructions fail before release AI Instruction Stress Test trigger misses, collisions, and behavior failures
turn one quality rule into an evaluator LLM Judge Prompt Builder a binary Pass/Fail judge prompt
check whether an AI judge matches humans LLM Judge Validator TPR, TNR, bias correction, and a trust verdict

Personal and creative workflows

You need to… Use You get
dial in espresso or V60 Coffee Brewing Coach exact grind, ratio, temperature, and time for the next brew
build a playlist that imports correctly Playlist Builder a verified playlist for your streaming service

Standalone applications

EarBrief — Turn Work Documents into Audio turns text and PDF reports into Chinese or English programs in a persistent mobile station. It replaces the retired listen-compare skill and runs independently of Claude or Codex. Install it using its own deployment guide; it is not a Skills CLI package. The current application has public listening and manual self-hosting setup.

Why these skills are different

  • Each skill solves one recognizable problem and produces a named result.
  • Destructive actions, uncertain claims, and unverified completion are called out explicitly.
  • Every released skill has its own product page, source package, direct install command, and release provenance.
  • Names stay plain for users while stable repository slugs keep existing installs working.

Source and releases

This repository is the source registry. Each directory contains one skill; products.json holds its public title, description, and GitHub topics; scripts/publish-skill releases committed source to github.com/m1nga/<skill-name> and verifies direct discovery through the Skills CLI.

scripts/verify-products
scripts/publish-skill diagnose-project-rebuild

Built by Ming. MIT licensed.

About

25 practical agent skills for Codex, Claude Code, and Cursor: plan products, recover projects, test prompts, preserve voice, and verify AI work.

Topics

Resources

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages