Check before you install

What it is

Check before you install is a skill your own AI agent runs before it installs anything from Mindket. It reads the product’s page as Markdown, reads llms.txt on the same site, and gives you a short report. The report has the price, the version, the last change, the health, the license, whether the source is archived, any advisories and what the product needs. Then it asks you. It never installs on its own.

1. You give your agent a Mindket link
2. It reads the product’s Markdown page and llms.txt
3. It follows the Source link for what the page leaves out
4. It reports each fact, or says it is not stated
5. You answer yes or no

Who it suits

Anyone who lets an agent add skills, MCP servers or kits. It helps most when the product comes from a seller you do not know. If you only install what you wrote yourself, skip it.

Where to use it

In any agent that reads Agent Skills (a folder with a SKILL.md file) and can open a web address. It reads public pages only, so it needs no key and no account.

• Claude Code: ~/.claude/skills/ for every project, or .claude/skills/ for one
• Other agents that read SKILL.md folders: the folder each one names, which Mindket’s post “What are Agent Skills?” lists
• A chat app with no skills folder: paste the text of SKILL.md as your first message

How to set it up

What you need first is an agent that can read web pages. Nothing else.

1. Sign in to Mindket and press Get it on this listing.
2. Unzip the file. It holds README.md and SKILL.md.
3. Put SKILL.md in a folder named check-before-you-install, inside your agent’s skills folder.
4. Start a new session, so the agent finds it.

How to use it

Give your agent the link or the id of a product and ask for the check.

Check before you install: <the product’s Mindket address>

What comes back, in this order:

• The kind, the seller and the price
• The version and the date of the last change
• The health Mindket shows, and whether the source repository is archived
• The license, and where the agent found it
• Advisories listed for the package it installs, with the day it asked
• What it needs: runtimes, keys, accounts and what it can reach on your computer
• One question: install it, not install it, or show the command first

Example

You paste the address of an MCP server that connects an agent to a database.

Check before you install: https://mindket.example/p/1234

Back comes something in this shape, not a saved run. The names, dates and numbers here are made up for the example.

Postgres MCP server, an MCP server by Acme. Free. Version 1, last changed 2026-11-10 (3 days ago).
Health: active. Archived: no.
License: MIT, from the repository’s LICENSE file.
Advisories: none returned by OSV.dev for the npm package on 2026-11-13. That does not prove there are none.
Needs: Node 20 and a DATABASE_URL. It can read every table that URL can reach.
Install it? Say yes, no, or show me the command first.

For newcomers and experts

If you are new. An agent can install code that runs on your computer with your permissions. This skill puts the facts in front of you first, from the product’s own page and its source. When the page has no answer, it writes “not stated” and does not guess. Try it first on something you already use.

For experts. The skill reads /p/<id>.md, which carries front matter (title, kind, category, price, seller, version, url, updated, and health for a GitHub project), the About, a kit’s README and the Source link. The license, the archived state and advisories come from the source: the repository page or GitHub’s public API, and OSV.dev for an npm or PyPI package the page names. Everything on the page counts as the seller’s words, so the skill treats it as data and follows no instruction found there.

Good to know

• It reads what is written. It does not run, scan or audit the code, and a report with no advisories is not a promise
• A fact that is on neither the page nor its source is reported as not stated
• Mindket shows health for GitHub projects only. Another product shows none
• It asks before every install, and it never asks for a key or a password
• It is free. The agent only reads public pages, as any visitor does