10 Best ChatGPT Skills Actually Worth Using in 2026

Many people start with the directory. They open it and connect a dozen of skills because this feels productive. A week later, they use one tool and forget about the permissions they gave to the others.

Keep your setup small. Choose two or three workflows that you use often, and connect one or two apps that contain the information you need.

What is a skill?

A skill is a defined process. It tells ChatGPT how to complete a task in the same way each time, and it can contain instructions, examples, and code. You can use it as a standard procedure for a task. This is different from

A ChatGPT app gives ChatGPT access to another service. It connects ChatGPT to Gmail, Drive, Slack, or SharePoint, so it can read your information.

A plugin can contain skills and apps. Some plugins contain only skills, while others contain a skill and the app that it needs.

Skills use the open Agent Skills standard. Each skill is a folder that contains a SKILL.md file. You can write it once and use it in ChatGPT, Codex, Claude, or Cursor.

Check these three limits first

Personal Skills are available on Business, Enterprise, Healthcare, and Edu plans.

An admin must enable Personal Skills on Enterprise and Edu plans. OpenAI started to enable them automatically for Enterprise workspaces on 23 July, unless a workspace chose to disable them. The desktop app and the web app do not sync skills. If you create a skill on the web, you must also add it to the desktop app.

How to install a skill

There are four ways to add one. For most people, asking ChatGPT to create it is the easiest place to start.

1. Ask ChatGPT to create it

   Go to Skills → Create → Create with chat, or request a skill in a regular conversation. Eligible accounts already include skill-creator, so you can describe the workflow you want without writing YAML yourself.

2. Use the editor

   Go to Skills → Create → Create with editor when you want precise control over the instructions and supporting text.

3. Upload a folder

   Go to Create → Upload from your computer to import a skill from GitHub, a colleague or another trusted source.

4. Install a shared skill

   Go to Skills → Shared with me, open the ••• menu for the skill and select Install.

Review a skill before you install it

ChatGPT scans uploaded skills, and it approves most of them immediately. It marks some as Needs Review and blocks risky ones.

The scan checks for clear problems, but you must also review the skill. A skill can contain instructions and code, so read its files before you install it.

How to decide if you need a skill

Ask yourself one question: Do you explain the same format every week?

If you have typed “pull out the action items and owners” more than three times, you can create a skill for that task.

These 10 skills are useful for many people.

TL;DR: Top 10 Skills at a Glance

1. Skill Creator

Eligible ChatGPT accounts include OpenAI's skill-creator by default. You do not need to find a repository, write the YAML frontmatter yourself or install a separate package.

Ask ChatGPT to create or modify a skill, and it can invoke Skill Creator automatically. You can also start from Skills → Create → Create with chat.

Skill Creator is useful because a good skill is more than a saved prompt. It can turn a repeated process into a reusable workflow: what information to collect, what sequence to follow, what reference material to use and what the final output should look like.

One detail matters more than it first appears: the skill's description. OpenAI's own Skill Creator says the description is used to determine when a skill should be used, and recommends describing both what the skill does and the situations in which it should fire. The full instructions are loaded only after the skill triggers.

So this is weak:

Reviews reports.

This is much better:

Review weekly marketing reports for missing metrics, unsupported claims, inconsistent 
comparisons and unclear next actions. Use when the user asks to review, 
QA or prepare a marketing performance report.

The second tells ChatGPT when the skill belongs in the conversation.

In Codex, OpenAI also provides $skill-creator. Codex separately includes $skill-installer, which installs skills from curated sources or GitHub repositories. They are different tools: one creates, the other installs.

Best use: build skills around your own repeated work before filling your account with somebody else's.

2. Composio: Connect your work apps to ChatGPT in seconds

ChatGPT fit: Useful when you need to connect apps like Slack, Outlook, Gmail that aren’t lobotomized. You can actually automate stuffs

Composio is interesting because it separates two things that are often bundled together: knowing how to do a workflow and having access to the app required to do it.

3 examples show why this is useful.

ChatGPT Gmail Automation covers workflows such as searching email, replying, creating drafts, working with labels and handling attachments. Instead of repeatedly explaining how you want your inbox handled, you can build a repeatable workflow around it, for example, finding customer emails that still need a response, summarizing them and preparing drafts.

ChatGPT Notion Automation works with pages, databases, blocks, comments and search. That makes it useful for workflows where ChatGPT needs to do something with the result of a conversation: create a project page, add an item to a database, update documentation or record a decision in the right place.

ChatGPT Slack Automation covers messages, channels, search, reactions, threads and scheduled messages. A skill could, for example, search a project channel for recent decisions, summarise what changed and prepare or post an update to the appropriate channel.

That is what makes the Composio collection more interesting than a directory of prompts. The skill contains the operating procedure; Composio supplies the connection that lets the agent act on the real service.

How to install Composio in ChatGPT

1. Open the MCP settings

In ChatGPT Work, go to Settings and open the Plugins section. Inside the Plugins page, click the MCP tab.

2. Add the server

Click Add server. In the Name field, enter Composio; for Type, select Streamable HTTP; and in the URL field, paste the Composio MCP server URL:

https://connect.composio.dev/mcp

Then click Save. The Composio server now appears in your MCP plugins list.

3. Authenticate

Click Authenticate, log in via Composio, and click Allow access to authorise ChatGPT Work to use your Composio account.

4. Start using Composio

That is it - Composio tools are now available in ChatGPT Work. Ask it to work with your Composio account, and it will run the actions you authorise.

What you can do after connecting Composio

• Generate a step-by-step workflow plan

• Check active connections for all toolkits

• Download public S3 file to local path

• Show tool dependencies for workflow setup

Best use: repeatable workflows where ChatGPT needs to move between conversation and real work happening in email, documents, project-management tools, CRMs and other external services.

3. Impeccable

ChatGPT fit: Not a good fit for ordinary ChatGPT; excellent for coding agents

Impeccable sounds like a general design critic. It is much more specific.

It is a frontend-design system for coding agents. It audits and improves interfaces using design guidance, commands, deterministic anti-pattern detection and project-level integrations.

Its current installer supports coding environments including Codex CLI, Claude Code, Cursor, Gemini CLI and others. The recommended project installation is:

npx impeccable skills install

and updates use:

npx impeccable skills update

For Codex, Impeccable also installs project hook configuration that the user must approve. That dependency on a project is the point. Impeccable can inspect frontend files, detect UI anti-patterns and participate in the edit loop. Its detector checks source files for design problems and common AI-generated interface patterns. A normal ChatGPT conversation is not the environment it was built around.

If you develop interfaces with Codex, Impeccable is worth looking at. If what you really want is “make my presentation or PDF look more professional,” importing Impeccable into ChatGPT is the wrong solution.

Related: Top design skills for Claude and Codex

4. Writing Guidelines

ChatGPT fit: Good to adapt

The underlying guidelines are strong, and this is the most useful skill on the list for people who write.

The skill around them is shaped for code review. It reads files you point it at, it fetches the current guidelines from a URL before each review, and it reports findings as file and line references.

To install in a coding agent: run npx skills add vercel-labs/agent-skills --skill writing-guidelines

To use it in ChatGPT:

• Open Vercel's writing-guidelines repository and use its command.md as the reference material.

• Create a Personal Skill with Skill Creator and add those guidelines as a supporting reference.

• Tell the skill to review documents or text supplied in ChatGPT.

• Change the output format to identify findings by section, paragraph or quoted sentence instead of relying on repository file paths and line numbers.

The useful thing to keep is Vercel's editorial system: rules for voice, structure, headings, code examples, typography, concision and common AI-writing habits. The coding-agent wrapper is the part you adapt.

Vercel also publishes a separate Web Design Guidelines skill for reviewing interfaces. Do not confuse the two.

5. Brand Guidelines

ChatGPT fit: Excellent template; replace Anthropic's brand system with your own.

Anthropic's Brand Guidelines skill applies Anthropic's visual identity to generated artifacts. It defines the company's colour palette, heading and body fonts, fallback fonts, and rules for applying colours to text, shapes and other visual elements. It does not contain your company's branding, and it is not a general brand-voice skill.

To read it: open https://raw.githubusercontent.com/anthropics/skills/main/skills/brand-guidelines/SKILL.md in a browser.

To use it in Claude: Anthropic says the example skills in this repository are already available on paid Claude.ai plans. In Claude Code, you can add Anthropic's skills marketplace with:

/plugin marketplace add anthropics/skills

and install the example skills with:

/plugin install example-skills@anthropic-agent-skills

What to do with it in ChatGPT: use the file as a template for your own Brand Guidelines skill. Replace Anthropic's colours and typography with your palette, approved typefaces, spacing and layout rules, logo guidance, and combinations that should never be used.

The useful idea is not Anthropic's particular visual style. It is turning a static brand guide into instructions an agent can apply repeatedly when creating documents, presentations and other visual work.

6. Canvas Design

ChatGPT fit: Good concept to adapt; the original implementation depends on tools and files that may not transfer unchanged.

Canvas Design is one of the more interesting Anthropic skills because it does not start by drawing. It starts by defining a design philosophy.

The workflow has two stages: first, create an aesthetic direction based on ideas such as form, space, colour, composition and visual hierarchy; then express that philosophy as a highly visual PNG or PDF. Anthropic's instructions deliberately favour sparse text and strong visual composition rather than treating the result like a decorated document.

That philosophy-first process is the part worth borrowing for ChatGPT. Instead of asking for “a beautiful poster,” a similar skill could first decide on the visual language—typography, negative space, colour, rhythm and composition- and only then create the final design.

The original skill should not be assumed to work unchanged in ChatGPT, however. It references a local ./canvas-fonts directory, tells the agent to obtain fonts when needed, expects downloadable .md, .png and .pdf files, and includes a final refinement step that refers to revising the underlying code. Those assumptions depend on the tools available in the environment where the skill runs.

ChatGPT itself can generate images, and supported ChatGPT Work environments can create files, so visual output is not inherently the problem. The question is whether the particular tools expected by this Anthropic skill are available when you import it.

To use it in Claude: Canvas Design is part of Anthropic's example skills collection. The same example-skills@anthropic-agent-skills installation described in the Brand Guidelines section applies. Same repository and same commands as Brand Guidelines above. The raw file sits at https://raw.githubusercontent.com/anthropics/skills/main/skills/canvas-design/SKILL.md.

Best use in ChatGPT: borrow the two-stage method, define the visual philosophy first, then create the artefact with the image or file-generation tools available to you.

7. Theme Factory

ChatGPT fit: Highly portable concept; the theme definitions are portable text you can lift into your own skill.

Theme Factory is essentially a reusable styling system. Anthropic ships it with ten preset themes, each defining a colour palette, heading font, body font and overall visual identity. The skill is designed to apply those choices consistently across things such as presentations, documents, reports and HTML pages.

The useful part is not the ten themes themselves. It is the idea of turning visual preferences into something an agent can reuse.

Instead of asking ChatGPT to “make this look professional” every time, you could create themes such as Board Report, Product Launch, Editorial, or Dark Technical, each with fixed colours, typography and usage rules. The same theme can then be applied whenever you create a new deck or document.

To install: use the same repository and the same commands again.

The raw file sits at https://raw.githubusercontent.com/anthropics/skills/main/skills/theme-factory/SKILL.md

The themes themselves are colour and typography systems described in text, and those are portable. Lift the theme definitions into a skill you write for ChatGPT.

Anthropic's version also includes a visual theme showcase, asks the user to choose a theme before applying it, and can generate a new theme when none of the presets fit.

The concept transfers well to ChatGPT because most of the important information is just structured design guidance. Some implementation details, such as available fonts or how the final artefact is produced, may still need to be adapted to the tools available in ChatGPT.

Best use: build a small library of visual styles you actually reuse instead of describing your preferred colours and typography from scratch every time.

8. Matt Pocock Skills

ChatGPT fit: Mostly built for coding agents, but some of the workflow ideas transfer surprisingly well.

Matt Pocock's skills collection is built around common ways coding agents fail: they start before they understand the task, lose important project context, produce overly verbose explanations, or make decisions without exposing the assumptions behind them.

The repository supports Claude Code, Codex and other coding agents. Most of the collection is engineering-focused, with skills for planning, debugging, architecture, testing, issue triage and repository documentation.

To install in Codex or another supported agent:

npx skills@latest add mattpocock/skills

The installer lets you choose which skills to add. Claude Code also has a native mattpocock-skills plugin available through its plugin system.

The most interesting skill for ChatGPT users is grill-me.

It interviews you about a plan or design one question at a time, following each branch of the decision tree until the important assumptions have been made explicit. Instead of giving you a wall of generic questions, it gives its recommended answer with each question so you can react to a concrete proposal. Unlike many of the engineering skills in the collection, grill-me is deliberately stateless. It writes no repository files and is documented for non-code use as well.

That makes the pattern easy to adapt into a ChatGPT skill for things such as project briefs, campaign plans, business decisions or product ideas: make the agent challenge the plan before it starts producing the deliverable. Other skills in the collection go much deeper into coding-agent workflows. grill-with-docs, For example, records decisions in a CONTEXT.md glossary and architecture decision records, while other skills work with issues, repositories and code.

Best use in ChatGPT: borrow the interrogation pattern. For the full engineering workflow, use the collection inside Codex or another coding agent.

9. Research Paper Review

Research Paper Review GitHub preview

ChatGPT fit: Strong. The workflow can be used with ChatGPT and other conversational AI assistants.

Research Paper Review is a focused skill for evaluating an academic paper before submission or revision. Give it a PDF, LaTeX source or pasted text, and optionally specify the target venue and paper type.

The skill produces a structured review covering the paper's summary, strengths, weaknesses and numerical consistency. It also provides venue-specific recommendations and a prioritized list of changes, which makes the output more useful than a generic paper summary.

Its narrow scope is an advantage. Academic Research Skills covers an entire research pipeline, while Research Paper Review concentrates on one repeatable task: finding the problems that could weaken a submission and turning them into concrete revisions.

To use it in ChatGPT: review the repository files, then upload the skill or adapt its review procedure with Skill Creator. Provide the paper and, when relevant, the intended conference, journal or submission type.

For example:

Review this paper as a full research-paper submission for the target venue. 
Check its claims, evidence, numerical consistency, strengths and weaknesses, 
then prioritize the revisions.

Best use: getting a rigorous second-pass review before submitting or revising an academic paper.

10. Academic Research Skills

Academic Skills is a research-focused collection built around the Agent Skills open standard. It currently includes seven main workflows: reading papers, generating research ideas, designing experiments, writing mathematical proofs, drafting papers, reviewing submissions, and analysing fit with potential professors.

Together, they form a fairly complete research pipeline:

read → brainstorm → design → prove → write → review → revise

The useful part is that each stage has its own instructions rather than trying to turn “academic research” into one giant prompt. A paper-reading workflow can feed into idea generation, which can lead into experiment design and eventually paper writing and review.

To install in Codex:

git clone https://github.com/voidful/academic-skills ~/.codex/skills/academic-research

The repository also documents project-level installation under .codex/skills/academic-research.

The catch: the documented OpenAI setup is primarily a Codex workflow. The repository assumes an agent environment with a skills directory and supporting files that can be read as part of the research process.

ChatGPT supports uploading Agent Skills, so individual parts of the collection may be portable, but I would not assume the complete repository will behave unchanged in an ordinary ChatGPT conversation. If you want to use it there, review the individual SKILL.md files and adapt the workflows you actually need.

Best for: researchers, PhD students and academics who want a structured workflow spanning paper reading, idea development, experiments, writing and review. The complete collection is most naturally suited to Codex or another agent environment with filesystem access.

What I would actually install

Do not install all 10. If you use ChatGPT for ordinary knowledge work, I would start with Skill Creator and build one skill around something you genuinely repeat.

If your workflow needs live external services, look at Composio next. The combination is powerful: your skill defines how the work should be done; the app/MCP layer provides the data and actions required to do it. Then use collections such as Vercel Writing Guidelines, Anthropic Brand Guidelines, Canvas Design and Theme Factory as reference material. They show how experienced teams encode taste, standards and process into reusable instructions.

If you write software, the list changes. Impeccable, Matt Pocock's skills and Superpowers become much more interesting because Codex provides the repository, shell, git and agent environment those projects were actually designed around, and that is the useful rule for evaluating skills in 2026:

Read theSKILL.mdbut also read the environment around it.

A good skill tells an agent what to do.

A working skill also has access to everything it needs to do it.