Stop prompting from scratch.Start from what works.
Featured building blocks
From need to useful, without the leap of faith.
01
Search by outcome
Describe the job, not the file format. Results rank on whether the thing actually does what you asked, and compare across platforms and models before you commit.
02
Read what it touches
Contents, permissions, requirements, licence and update history are separate fields on every listing — not paragraphs in a README you have to trust.
03
Put it to work
One command, the exact version you inspected, and updates that arrive with notes from the person who wrote them.
Trust is product information.
Most catalogues bury what a tool actually does inside a paragraph of marketing. We break compatibility, permissions, requirements, licence and safety status into separate fields so you can judge a product before it touches your workflow.
A good AI product should tell you what it does, what it touches, and why you can trust it.
Human-readable permissions
Every listing states what it reads, writes and calls out to — in a sentence, not a config file.
Visible update history
Every version, with notes from the person who shipped it. Pin the one you inspected.
Licence on the label
Use, redistribution and attribution are clear before you install anything.
One useful release every Friday.
What shipped, what got reviewed, and what is worth your attention. No launch announcements.
One email a week. Unsubscribe any time.