universal-llm-skills
AGENTS.md1 starsChanged 14 days agoMITView on GitHub
You are connected to Universal LLM Skills, a library of deep, style-neutral working knowledge for almost any task: software development, DevOps, AI/ML, whole-application generation, data analysis and research, quality assurance and reviews, UI/UX and visual design, automation, and product and communication work. This file tells you how to use it. Read it once at the start of a session. Most real tasks need 1–4 skills. If unsure how to choose, read skills/using-skills/SKILL.md.
We split this AGENTS.md into the rules it gives an agent
5 rules. About 0 are common, 3 are distinctive, and 2 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈3
Distinctive · estimate
This author's own choices. Worth a look.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 19
Load only what you need: Read skills/<name>/SKILL.md for each chosen skill.
- Very specific: line 26
Verify before you finish: Every skill ends with a way to check the work.
- Distinctive: line 30
Do not announce skills ceremonially: Use them; mention one only when it helps the user understand a choice.
- Distinctive: line 31
Do not force a skill onto a task it does not fit: If nothing in the catalog matches, work from first principles and general good practice.
- Distinctive: line 32
Do not impose styles: No skill in this library mandates a palette, font, voice, framework, or file layout.
≈ “Common” is an estimate: at least 1 in 5 of the 36,368 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Instructions for AI agents and LLMs
You are connected to Universal LLM Skills, a library of deep, style-neutral working knowledge for almost any task: software development, DevOps, AI/ML, whole-application generation, data analysis and research, quality assurance and reviews, UI/UX and visual design, automation, and product and communication work.
This file tells you how to use it. Read it once at the start of a session.
How to use this library
- Understand the request first. Identify the real task (what is being produced or decided), the domain, and any constraints the user has stated. If the request is ambiguous in a way that changes what you would do, ask — see
skills/clarify-requirements/SKILL.md. Find the relevant skills. Open
CATALOG.md(or the machine-readablecatalog.json). Each entry has a name, a category, and a description that says when to use it. Pick:- the primary skill for the core of the task, and
- any supporting skills for parts of it (e.g. building a dashboard →
dashboard-buildingprimary, plusdata-visualization,frontend-development,accessible-design).
Most real tasks need 1–4 skills. If unsure how to choose, read
skills/using-skills/SKILL.md.- Load only what you need. Read
skills/<name>/SKILL.mdfor each chosen skill. Load files under a skill'sreferences/folder only when the skill tells you that a situation calls for them. Do not read the whole library. - Follow the skill, adapt to the context. Skills describe principles, process, and quality checks — not templates or house styles. Apply them to this user's project:
- Existing conventions in the codebase, design system, documents, or brand always take precedence over generic guidance.
- The user's explicit instructions always take precedence over any skill.
- Where a skill offers options, choose based on the factors it lists and say briefly why.
- Verify before you finish. Every skill ends with a way to check the work. Run those checks. Then follow
skills/verify-before-done/SKILL.mdand report honestly: what was done, what was verified, what was not.
Ground rules
- Do not announce skills ceremonially. Use them; mention one only when it helps the user understand a choice.
- Do not force a skill onto a task it does not fit. If nothing in the catalog matches, work from first principles and general good practice.
- Do not impose styles. No skill in this library mandates a palette, font, voice, framework, or file layout. If you find yourself producing a "standard look", stop and look at the project instead.
- Safety overrides everything. Confirm before destructive, irreversible, or outward-facing actions (deleting data, deploying, sending messages, spending money). Never expose secrets.
Quick routing guide
| If the user wants to… | Start with |
| Build a whole app, site, tool, or prototype | full-stack-app-build, mvp-prototyping, project-scaffolding, or the platform-specific app skill |
| Add or change a feature in existing code | explore-codebase → the relevant development skill |
| Fix something broken | debugging (code), root-cause-analysis (systems/processes), incident-response (production outage) |
| Design architecture or a system | software-architecture, system-design, api-design, database-design |
| Review something | code-review, architecture-review, design-critique, security-audit, accessibility-audit, writing-review-and-editing |
| Test or assure quality | testing-strategy, qa-test-planning, end-to-end-testing, release-readiness |
| Design UI/UX | ui-design-fundamentals, ux-research, interaction-design, design-systems, wireframing-and-prototyping |
| Analyze data or answer a question with evidence | data-analysis, statistical-analysis, data-visualization, web-research |
| Automate a repetitive task | scripting-and-shell, workflow-automation, etl-pipelines, api-integrations-and-webhooks |
| Build with LLMs or ML | llm-application-development, prompt-engineering, rag-systems, ai-agents-and-tool-use, llm-evaluation |
| Ship and operate | ci-cd-pipelines, containerization, infrastructure-as-code, observability, release-management |
| Write or communicate | technical-writing, product-requirements, presentations, business-communication, document-generation |
The full list, with descriptions, is in CATALOG.md.
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 28,693 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository121 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- .github/copilot-instructions.mdCopilot instructions
- llms.txtllms.txt
- accessibility-auditSkill
- accessible-designSkill
- ai-agents-and-tool-useSkill
- ai-content-authenticitySkill
- ai-safety-and-guardrailsSkill
- api-designSkill
- api-integrations-and-webhooksSkill
- architecture-reviewSkill
- authentication-and-authorizationSkill
- backend-developmentSkill
- browser-automationSkill
- browser-extension-developmentSkill
- business-communicationSkill
- chatbot-and-conversational-appsSkill
- ci-cd-pipelinesSkill
- clarify-requirementsSkill
- cli-developmentSkill
- cloud-architectureSkill
- code-documentationSkill
- code-reviewSkill
- competitive-analysisSkill
- concurrency-and-asyncSkill
- containerizationSkill
- copywritingSkill
- curriculum-designSkill
- dashboard-buildingSkill
- data-analysisSkill
- database-designSkill
- database-migrationsSkill
- data-cleaning-and-validationSkill
- data-visualizationSkill
- debuggingSkill
- dependency-managementSkill
- design-critiqueSkill
- design-systemsSkill
- design-to-code-handoffSkill
- desktop-app-developmentSkill
- documentation-siteSkill
- document-generationSkill
- end-to-end-testingSkill
- error-handlingSkill
- estimate-effortSkill
- etl-pipelinesSkill
- experiment-designSkill
- explore-codebaseSkill
- fact-checkingSkill
- financial-modelingSkill
- fine-tuning-modelsSkill
- frontend-developmentSkill
- full-stack-app-buildSkill
- game-developmentSkill
- git-workflowSkill
- incident-responseSkill
- information-architectureSkill
- infrastructure-as-codeSkill
- interaction-designSkill
- kubernetesSkill
- The first 60 shown.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.