Safety / Methodology
How we review
Every mod on this site goes through two layers: a machine scan that maps what the code can do, and human judgment on whether you should trust it. Here's exactly what each layer does — and what it doesn't.
01 /
The machine scan
Every week, our scanner clones each mod's repository, finds every mod inside it,
and statically reads which Claude Code APIs the code calls — $.fs.read,
$.http.fetch, $.env.get, and so on. It never installs or
runs the code. It only reads it.
The calls are graded into a reach level:
| Level | Name | Meaning |
|---|---|---|
| 0 | Draws and remembers | Draws UI, keeps local state, plays audio. Can't touch your files or the network. |
| 1 | Reads | Reads files, env vars, settings, your transcript. Can see things, can't change or send them. |
| 2 | Writes or runs | Writes files, runs processes, changes config, drives Claude. Can change things on your machine. |
| 3 | Network | Makes network requests or calls MCP servers. Can send data out. |
02 /
What the machine can't tell you
- We read code, we don't run it. Static analysis sees capabilities, not behavior. A mod can read your env vars; whether it does something bad with them is a different question.
- We can't see intent.
$.fs.readin a backup tool and$.fs.readin a sketchy mod look identical to the scanner. - "Needs review" is not an accusation. A flag means "a human should look at this", not "this mod is malware."
03 /
The human review
For the mods that matter most — the popular ones, the powerful ones, the ones with broad permissions — a human reads the source, checks what the machine found, and writes a one-line verdict. A verdict answers three things: what it does, what it can touch, and whether you should install it.
Each reviewed mod gets a badge: SAFE (green), CAUTION (amber), or RISK (red). The badge starts from the machine's reach level; a reviewer can override it with a written reason.
04 /
How to read a mod page
- Badge — our judgment: SAFE / CAUTION / RISK.
- Reach labels — plain-English chips of what the code can do:
reads files,network,runs processes. - Can see — which parts of your Claude Code session the mod observes.
- API calls — the raw list, for people who read code.
- Validation — whether the mod passes Claude Code's own plugin validator, and on which version.
05 /
How fresh is this?
The machine rescan runs weekly; badges and capability reports update automatically. Human reviews show the Claude Code version they were reviewed against. A review doesn't expire, but the code may have changed since — the "last scanned" date on each mod page tells you how fresh the machine data is.
06 /
Where the catalog comes from
We track 4,296+ Claude Code mods across GitHub — community plugins, marketplaces, and independent repos. Our discovery pipeline finds new ones weekly, and every new find is automatically scanned and badged before it appears here.