Security scanning

What the scan looks for, and what it does not prove.

What we look for

Suspicious shell commands: recursive deletes, privilege escalation, persistence via cron or launch agents, history tampering.
Credential and secret access: SSH keys, cloud credentials, .env files, keychains, browser profiles, hardcoded token patterns.
Hidden network activity: piping downloads into a shell, raw-IP hosts, reverse shells, paste and short-link sites, webhook exfiltration.
Encoded or obfuscated payloads: base64 executed at runtime, long encoded blobs, hex-escaped strings, zero-width characters.
Prompt-injection indicators: overriding prior instructions, hiding behaviour from the user, fake system authority, instructions to transmit secrets, disabled safety flags.

Statuses

Not scanned
Nothing has been scanned yet. New manual listings without a source get scanned on their first content change.
Security scanned
No indicators found by the current scan version. This is a heuristic result, not a guarantee.
Security verified
Scanned and additionally reviewed by a human.
Warning
Medium-severity indicators. Usually a legitimate command mentioned in prose; read the security section before installing.
High risk
High or critical indicators found. The skill is held for review and heavily discounted in ranking.
Blocked
Multiple critical indicators. Cannot be published or ranked.

What it can’t do

The scanner reads text. It can’t execute a skill, can’t follow every link, and can’t know what a model will do with an ambiguous instruction. Treat “Scanned” as “nothing obvious”, not “safe”. Read the source before you run anything with access to your machine or accounts, and report anything that behaves differently from its description.