PR #10 added the platform-postgres backup patch and its restore-drill
Cluster. Both hold barmanObjectStore.s3Credentials, where accessKeyId
and secretAccessKey are SecretKeySelectors (name and key). The
structured scan treats any non-string value under a key ending in
"access key" as a credential literal, so the source scan reported
CREDENTIAL_LITERAL for both files on main.
An entry directly under s3Credentials, named accessKeyId or
secretAccessKey, whose value is a mapping of exactly name and key
with non-empty strings, is now a reference. This mirrors the existing
exact rule for the Grafana client-secret secretKeyRef.
Rejected: a general "name and key is always a reference" rule. It
would also pass a literal hidden under some other credential key that
happens to have those two fields.
Risk: none beyond the two key names under that one parent. A literal
string, an extra child, or the same mapping under another parent is
still rejected (new test cases).
Verification: on desktop, the source scan of main 205927f reported the
two files before this change and passes after it. The unit test
passes in a non-canonical clone.