Universal Agent Skills
Production AI skills and tool protocols across major agent runtimes
Hardened instruction sets that eliminate hallucinations, enforce senior engineering discipline, scrape verified datasets, and run seamlessly across Claude Code, Cursor, Google Antigravity, ChatGPT, and Local LLMs.
Target Audience: Useful for anyone who wants to empower their AI agents or tools with production capabilities without technical friction.
- 01SkillPublic
Zero Hallucination Coder
CodingOriginalEnforces verified reasoning before any code is generated.
By Mehansh Barthwal•MIT•1 files - 02SkillPublic
Universal Scraping Architect
ResearchOriginalDynamic scraping strategy router with token budgeting and checkpointing.
By Mehansh Barthwal•MIT•1 files - 03SkillPublic
Pain Point Miner
ResearchOriginalAuthentic customer complaint extraction with mechanical deduplication.
By Mehansh Barthwal•MIT•5 files - 04ToolPublic
ArXivist
ResearchTranslates scientific research papers into runnable code repositories.
By Mehansh Barthwal (Skill) and qosi org (Original Repository)•MIT•14 files - 05WorkflowPublic
Loop Until Done
OrchestrationOriginalAutonomous self evaluation loop rewriting drafts until rubric bars clear.
By Mehansh Barthwal•MIT•1 files - 06SkillPublic
JD to Job
CareerOriginalExtracts operational role requirements into tailored application materials.
By Mehansh Barthwal•MIT•2 files - 07SkillPublic
Resume Unrejectable
CareerOriginalFour stage evaluation gauntlet reflecting actual hiring pipelines.
By Mehansh Barthwal•MIT•1 files - 08SkillPublic
Locality Delivery Scraper
ResearchOriginalMethodology playbook for mapping urban food delivery markets.
By Mehansh Barthwal•MIT•1 files - 09ToolPublic
IPYNB Editor
CodingOriginalProgrammatic cell manipulation preserving notebook JSON integrity.
By Mehansh Barthwal•MIT•2 files - 10SkillPublic
Markdown Converter Router
ResearchOriginalIntelligent format routing across AnyDoc and MarkItDown.
By Mehansh Barthwal•MIT•1 files - 11SkillPublic
Humanizer
WritingStrips robotic hallmarks and promotional padding from machine prose.
By Siqi Chen•MIT•6 files - 12SkillPublic
Stop Slop
WritingCompact prompt rules that block formulaic filler during initial generation.
By Hardik Pandya•MIT•7 files - 13SkillPublic
AI Watermarks Remover
WritingCombines zero width character scrubbing with stylistic rephrasing.
By Mehansh Barthwal / Community•MIT•1 files - 14SkillPublic
Ponytail
CodingSenior developer minimalism prioritizing standard libraries and YAGNI.
By DietrichGebert•MIT•122 files - 15ToolPublic
Graphify
ArchitectureConstructs queryable knowledge graphs from codebases and documents.
By Safi Shamsi•Apache-2.0•766 files - 16ToolPublic
GSD Core
OrchestrationPhase based autonomous development with user acceptance gates.
By Open GSD•MIT•2189 files - 17ToolPublic
Agent Reach
ResearchUnified research interface and Model Context Protocol across fifteen networks.
By Agent Eyes•MIT•108 files - 18ToolPublic
Agent Video
ResearchTransforms video recordings into structured frames and aligned transcripts.
By Bradley Bonanno•MIT•27 files - 19ToolPublic
Frontend Slides
DesignBuilds animation rich HTML slide decks or converts from PowerPoint.
By Zara Zhang•MIT•160 files - 20ToolPublic
MarkItDown
ResearchMicrosoft utility converting diverse file formats into Markdown.
By Microsoft Corporation•MIT•158 files - 21ToolPublic
Ralph
OrchestrationConverts requirements specifications into deterministic agent code.
By snarktank•MIT•27 files - 22ToolPublic
SkillOpt
OrchestrationOffline agent review cycle consolidating recurring session corrections.
By Microsoft Corporation•MIT•368 files - 23SkillPublic
LLM Council
OrchestrationFive advisor debate protocol peer reviewing proposals anonymously.
By Andrej Karpathy (Concept) / Mehansh Barthwal (Protocol)•MIT•2 files - 24SkillPublic
Karpathy Guidelines
CodingFour always on instincts preventing common agent coding failures.
By Andrej Karpathy (Observations) / Mehansh Barthwal (Distillation)•MIT•1 files - 25SkillPublic
Taste Skill
DesignAnti slop frontend design direction for web applications.
By Community Design Collective•MIT•15 files - 26SkillPublic
Truth Prompt
ResearchStrictly separates verified facts from inferences and assumptions.
By Epistemic AI Research•MIT•3 files - 27SkillPublic
OpenHuman Context
ArchitectureArchitecture context for the tinyhumansai/openhuman codebase.
By tinyhumansai•MIT•12 files - 28SkillInternal
Zapier MCP Config
ArchitectureReference configuration for Zapier Model Context Protocol integrations.
By Mehansh Barthwal•Private•1 files - 29SkillInternal
Knowledge Base
ArchitectureEngineering playbook and deployment patterns to avoid repeating past errors.
By Mehansh Barthwal•Internal•1 files - 30WriteupInternal
On Building the Catalog
ArchitectureOriginalWhy this collection maintains a machine readable registry.
By Mehansh Barthwal•MIT - 31WriteupInternal
Portability Across Runtimes
ArchitectureOriginalHow one skill format operates across six agent runtimes.
By Mehansh Barthwal•MIT - 32WriteupInternal
Skills, Not Prompts
ArchitectureOriginalThe essential difference lies in triggers and negative constraints.
By Mehansh Barthwal•MIT - 33WriteupInternal
Notes on Anti Slop Engineering
WritingOriginalApplying intentional de-biasing to prose and user interfaces.
By Mehansh Barthwal•MIT
Ecosystem Architecture
Designed for cross runtime reliability without vendor lock in.
Each skill in this repository is built as a self contained specification declaring trigger conditions, negative constraints, and output verification rubrics. This ensures predictable execution even when models confront tasks outside their initial training distribution.
External tools and upstream inspirations are credited openly in the attribution matrix. Items marked internal represent private runbooks and personal configurations.