You are reading the tutorials for Cordon 0.4 (0.4.1). Read this for 0.7, the current release.
Reachability, cut noise without hiding anything
The loudest complaint about scanners: a CVE in a package three levels down that your code never touches, screaming at the same volume as one you call constantly.
cordon-scanner scan . --reachabilityANNOTATE, NEVER SUPPRESS ┌─────────────────────────────────────────────────────────────┐ │ An unreached finding is LOWERED and TAGGED — never dropped.│ │ Gate on "reached only" if you want; the rest stay in the │ │ report where a reviewer can still see them. │ └─────────────────────────────────────────────────────────────┘
vulnerable dependency
│
▼
does first-party code import it?
│
┌──────┴───────┐
YES NO
│ │
│ is it a DIRECT dependency?
│ │ │
│ YES NO (transitive)
│ │ │
▼ ▼ ▼
IMPORTED UNKNOWN NOT_IMPORTED
(severity (severity (severity LOWERED
kept) kept) one step + tagged)UNKNOWN ≠ UNREACHED ─────────────────── A direct dep you don't see imported may be loaded dynamically, or under a name the distribution doesn't publish (beautifulsoup4 → bs4). Import-checking can't rule that out, so it stays UNKNOWN and its severity does NOT move. MALWARE never moves ─────────────────── A compromised release must not be installed whether or not it is called. Reachability only ever lowers a VULNERABILITY, never a malicious package.
BEFORE AFTER --reachability ────── ────────────────────── HIGH CVE in leftpad (transitive) ──▶ MEDIUM CVE in leftpad "…not imported by first-party code, lowered rather than dropped" HIGH CVE in requests (you import) ──▶ HIGH CVE in requests "…imported by first-party code"
This is the cheap, always-available import tier. The precise call-graph tier, is the vulnerable symbol on a path you actually reach, builds on the AST providers (tutorial 13) and is a later layer.