Summary: ChatGPT skills are saved workflows — a folder with a SKILL.md file that ChatGPT loads whenever your request matches its description. They became generally available on July 9, 2026, for eligible Business, Enterprise, Healthcare and Edu plans, per a Firecrawl guide published October 30, 2026. Paired with the Firecrawl plugin, a skill can reach live web pages and structured data providers, which turns a lookup you repeat every week into one request.
What a skill actually is
A skill is small on purpose. The SKILL.md file carries a name, a description of when to use it, and numbered steps. ChatGPT only reads the name and description until a request matches — then it loads the full instructions. The smallest useful version looks like this:
---
name: weekly-update
description: Turn my notes into a weekly team update. Use when I paste notes and ask for an update, recap or status report.
---
1. Group the notes into Done, In progress and Blocked.
2. Write one line per item, starting with a verb.
3. End with the top three priorities for next week.You can call a skill by name with @ in the composer, or let ChatGPT trigger it automatically. Folders can also hold scripts and templates, and the format follows the open Agent Skills standard — the same one Codex uses, so one folder works in both tools.
Who can build them, and how
Skills live under Plugins in the sidebar. Business and Healthcare users can create, upload, share and install them; on Enterprise and Edu, admins control those permissions. Plus, Pro, Go and Free aren’t listed for creating your own — a gap Plus users have already raised on OpenAI’s community forum.
Four creation paths: chat (via the built-in @skill-creator, which drafts the file for you), the Skills editor, uploading a folder, or installing one a teammate shared. If you’re drafting the description yourself, the guidance is consistent: front-load the trigger words, name the tools to call, keep each skill to one job, and split big workflows into small combinable skills. The description does most of the work, since that’s all ChatGPT sees before deciding to load.
If you’ve used custom GPTs, note the distinction: a skill teaches one task, a project holds shared files and chats, and custom GPTs retire on December 11, 2026, migrating into plugins. Custom actions don’t carry over, so plan replacements.
Wiring in real data with Firecrawl
A skill only holds instructions — it needs an app to reach anything current. The Firecrawl plugin connects ChatGPT to Firecrawl’s search, scrape and crawl endpoints plus Firecrawl Alexandria’s provider network (SEC EDGAR, People Data Labs, Particle, Sleeper and more). Setup is short: add the plugin, sign in to Firecrawl, then name the plugin and provider inside your SKILL.md. Two caveats from the guide: Alexandria requires a signed-in Firecrawl account, and some providers like Apollo need a team admin to accept their terms once before the first call.
This pattern should feel familiar if you’ve read our earlier piece on why ChatGPT can’t crawl a site on its own — skills are the reusable wrapper around that same fix.
Three of the five example skills
The Firecrawl post ships five copy-ready skills. Three worth studying:
- Fantasy lineup check — reads a Sleeper league, the NFL.com injury report, and Draft Sharks rankings, then returns a start/sit table with a source per row.
- Account brief — takes a company domain, enriches it via People Data Labs, lists funding rounds from Fireround, finds target contacts through Apollo, and adds recent news — with a citation for every fact.
- Filing brief — pulls the two most recent 10-Ks or 10-Qs from SEC EDGAR, diffs risk factors and MD&A sections, and appends news since the filing date.
The common thread: each skill names its providers explicitly, specifies the output format, and states what it won’t do (no betting advice, no buy/sell guidance). That’s the real template — narrow trigger, named tools, structured output. Start with @skill-creator, describe your most-repeated lookup, and test it against a real request before sharing it with your team.