conformance/tests/l0-static/vuln-scanner-supply-chain/check.sh

@@ -16,6 +16,11 @@
 #       (6) the pod: non-root, read-only rootfs, drop ALL, seccomp RuntimeDefault, no host
 #           namespaces or hostPath, CPU and memory limits on every container, an ephemeral-storage
 #           limit on the scanner, and a size-limited /cache emptyDir.
+#       (8) a renovate.json5 rule covers every file under platform/vuln-scanner: automerge:false,
+#           minimumReleaseAge of at least 7 days, and no later rule turns auto-merge back on (the
+#           python image holds the cluster-read token);
+#       (9) the scanner names the private registry it cannot reach, so that registry reports
+#           credential-required rather than a permanent failed.
 #   AND (7) each evaluator is run on a mutated copy and must flag it (non-vacuity).
 # =============================================================================
 set -euo pipefail
@@ -150,6 +155,33 @@ def eval_renovate(rules):
     return bad
 
 
+SCANNER_FILES = "platform/vuln-scanner/**"
+
+
+def eval_renovate_files(rules):
+    positions = [i for i, r in enumerate(rules) if SCANNER_FILES in (r.get("matchFileNames") or [])]
+    if not positions:
+        return [f"no packageRule has matchFileNames {SCANNER_FILES}"]
+    bad = []
+    rule = rules[positions[-1]]
+    if rule.get("automerge") is not False:
+        bad.append("the platform/vuln-scanner file rule is not automerge:false")
+    age = VERSION_UPDATE_RULE_RE.match(rule.get("minimumReleaseAge") or "")
+    if not age or int(age.group(1)) < 7:
+        bad.append("the platform/vuln-scanner file rule has no minimumReleaseAge of at least 7 days")
+    if any(r.get("automerge") is True for r in rules[positions[-1] + 1:]):
+        bad.append("a rule after the platform/vuln-scanner file rule re-enables auto-merge")
+    return bad
+
+
+def eval_private_registries(pod):
+    scanner = next(c for c in pod["containers"] if c["name"] == "scanner")
+    env = {e["name"]: e.get("value") for e in scanner.get("env") or []}
+    if env.get("VULN_PRIVATE_REGISTRIES") != "gitea.${BASE_DOMAIN}":
+        return ["the scanner does not name the private registry gitea.${BASE_DOMAIN}"]
+    return []
+
+
 def eval_rbac(rbac):
     bad = []
     croles = [d for d in rbac if d.get("kind") == "ClusterRole"]
@@ -224,6 +256,10 @@ rules = renovate.get("packageRules") or []
 check(len(rules) > 5, "(4) parsed renovate.json5 packageRules (parser degenerated)")
 for v in eval_renovate(rules):
     check(False, f"(4) {v}")
+for v in eval_renovate_files(rules):
+    check(False, f"(8) {v}")
+for v in eval_private_registries(pod):
+    check(False, f"(9) {v}")
 
 rbac = docs(COMP / "rbac.yaml")
 for v in eval_rbac(rbac):
@@ -274,6 +310,23 @@ for name, fn in (("Trivy automerge flipped true", ren_auto), ("minimumReleaseAge
                  ("Trivy removed from Renovate rules", ren_drop), ("later auto-merge rule appended", ren_later)):
     check(bool(eval_renovate(mutated(fn, rules))), f"(7) Renovate evaluator missed mutant: {name}")
 
+# Renovate file-rule mutants
+def files_rule(r):
+    return next(x for x in r if SCANNER_FILES in (x.get("matchFileNames") or []))
+def files_auto(r): files_rule(r)["automerge"] = True
+def files_age(r): files_rule(r)["minimumReleaseAge"] = "1 day"
+def files_drop(r): r.remove(files_rule(r))
+def files_later(r): r.append({"matchDatasources": ["docker"], "automerge": True})
+for name, fn in (("file rule automerge true", files_auto), ("file rule age shortened", files_age),
+                 ("file rule removed", files_drop), ("later rule re-enables auto-merge", files_later)):
+    check(bool(eval_renovate_files(mutated(fn, rules))), f"(7) file-rule evaluator missed mutant: {name}")
+
+def reg_drop(p):
+    next(c for c in p["containers"] if c["name"] == "scanner")["env"][:] = [
+        e for e in next(c for c in p["containers"] if c["name"] == "scanner")["env"]
+        if e["name"] != "VULN_PRIVATE_REGISTRIES"]
+check(bool(eval_private_registries(mutated(reg_drop, pod))), "(7) private-registry evaluator missed the removal")
+
 # Pod mutants
 def pod_root(p): p["securityContext"]["runAsNonRoot"] = False
 def pod_host(p): p["hostNetwork"] = True

conformance/tests/l1-rendering/vuln-scanner-acceptances/check.py

@@ -94,6 +94,15 @@ def status(when, accepted, errs=()):
     return built, built["findings"][0]
 
 
+def reason_entry(length):
+    return acceptances.parse_text("- id: CVE-1\n  reason: " + "r" * length + "\n  expires: 2026-12-01\n", NOW.date())
+
+
+found, problems = reason_entry(3000)
+check("a reason over 2048 characters ignores the entry and reports it", found == [] and any("2048" in p for p in problems))
+found, problems = reason_entry(2048)
+check("mutation: a reason of exactly 2048 characters is valid", len(found) == 1 and problems == [])
+
 entry = {"id": "CVE-1", "image": None, "reason": "not reachable", "expires": date(2026, 11, 9)}
 built, finding = status(NOW, [entry])
 check("unexpired acceptance marks the finding accepted", finding["status"] == "accepted"

conformance/tests/l1-rendering/vuln-scanner-api-metrics/check.py

@@ -150,6 +150,10 @@ with tempfile.TemporaryDirectory() as tmp:
         mutate(document)
         status, _, _ = post(internal, document)
         check("triage rejects " + name, status == 400)
+    for name, url in (("an uppercase scheme", "HTTP://gitea.test/pulls/1"), ("an empty url", ""), ("a javascript url", "javascript:alert(1)")):
+        bad = json.loads(json.dumps(valid))
+        bad["remediation"][0]["url"] = url
+        check("triage rejects " + name, post(internal, bad)[0] == 400)
     check("triage rejects a wrong content type", post(internal, valid, "text/plain")[0] == 415)
     status, _, _ = request(internal, "POST", "/triage", b"x" * 300000, {"Content-Type": "application/json"})
     check("triage rejects an oversized body", status == 413)

conformance/tests/l1-rendering/vuln-scanner-page/check.py

@@ -170,6 +170,39 @@ _, _, body = render()
 check("default view excludes nothing", all(f["id"] in body for f in REPORT["findings"]))
 check("remediation column shows pr-open with a link", 'href="https://gitea.test/pulls/1"' in body and "pr-open" in body)
 
+# Status and state belong to the line they describe (R-VULN-04).
+images = [t for t in REPORT["targets"] if t["type"] == "image" and t["workloads"]]
+check("fixture has two image targets with workloads", len(images) >= 2)
+mixed = copy.deepcopy(REPORT)
+first, second = next(t for t in mixed["targets"] if t["id"] == images[0]["id"]), next(t for t in mixed["targets"] if t["id"] == images[1]["id"])
+first["state"], first["reason"] = "scanned", None
+second["state"], second["reason"] = "stale", "last success long ago"
+base = dict(REPORT["findings"][0], id="CVE-MIXED")
+base.pop("acceptance", None)
+mixed["findings"] = [
+    dict(base, targetId=first["id"], status="accepted", acceptance={"reason": "only-on-the-first", "expires": "2026-12-01"}),
+    dict(base, targetId=second["id"], status="fix-available"),
+]
+_, _, body = render(mixed, "view=cves&q=CVE-MIXED")
+check("a mixed CVE reads as mixed, with counts", "mixed: 1 fix-available, 1 accepted" in body)
+check("a mixed CVE's row does not carry one image's acceptance", "<td>mixed: 1 fix-available, 1 accepted</td>" in body)
+first_line = page.e(first["id"]) + " — target scanned — status accepted (only-on-the-first, until 2026-12-01)"
+second_line = page.e(second["id"]) + " — target stale (last success long ago) — status fix-available"
+check("the accepted image's line carries its own acceptance", first_line in body)
+check("the other image's line carries its own status and state", second_line in body)
+check("no line attaches the acceptance to the unaccepted status", "status fix-available (only-on-the-first" not in body)
+check("workload lines inside the expansion carry state and status",
+      "target stale (last success long ago) — status fix-available</li>" in body
+      and "containers: " + page.e(", ".join(second["workloads"][0]["containers"]) or "none") + " — target stale" in body)
+_, _, plain = render(query="view=cves&status=accepted")
+check("mutation: a uniformly accepted CVE does not read as mixed", "mixed:" not in plain and "CVE-2026-1003" in plain)
+_, _, body = render(mixed, "view=workloads")
+check("the workloads images column labels each image's state", page.e(second["id"]) + " — target stale (last success long ago)" in body)
+check("the workloads expansion labels each finding's status, image and state",
+      "status fix-available — " + page.e(second["id"]) + " — target stale (last success long ago)</li>" in body)
+check("the workloads expansion keeps an acceptance on its own line",
+      "status accepted (only-on-the-first, until 2026-12-01) — " + page.e(first["id"]) + " — target scanned</li>" in body)
+
 big = copy.deepcopy(REPORT)
 big["findings"] = [dict(REPORT["findings"][0], id="CVE-2026-" + format(i, "05d")) for i in range(250)]
 _, _, body = render(big, "view=cves&size=50&page=2")

conformance/tests/l1-rendering/vuln-scanner-scan-loop/check.py

@@ -95,14 +95,14 @@ class FakeRunner:
         pass
 
 
-def engine(tmp, runner=None, clock=None, acc="[]\n", cluster=CLUSTER):
+def engine(tmp, runner=None, clock=None, acc="[]\n", cluster=CLUSTER, private=()):
     path = os.path.join(tmp, "acceptances.yaml")
     if not os.path.exists(path):
         Path(path).write_text(acc)
     clock = clock or Clock()
     runner = runner or FakeRunner()
     runner.clock = clock
-    return main.Engine(tmp, FakeClient(cluster), runner, path, clock=clock, publish_seconds=0), runner, clock
+    return main.Engine(tmp, FakeClient(cluster), runner, path, clock=clock, publish_seconds=0, private_registries=private), runner, clock
 
 
 def targets(eng):
@@ -239,6 +239,184 @@ with tempfile.TemporaryDirectory() as tmp:
     thread.join(10)
     check("a crash in one cycle is logged and the loop continues", len(calls) == 2 and not thread.is_alive())
 
+class Killed(BaseException):
+    """What an OOM kill looks like from inside the process: no `except Exception` runs."""
+
+
+class KillingRunner(FakeRunner):
+    def __init__(self, victim, data_dir):
+        super().__init__()
+        self.victim = victim
+        self.data_dir = data_dir
+        self.on_disk = None
+
+    def scan(self, reference):
+        if reference == self.victim:
+            self.on_disk = json.loads(Path(self.data_dir, "state.json").read_text()).get("inFlight")
+            raise Killed()
+        return super().scan(reference)
+
+
+with tempfile.TemporaryDirectory() as tmp:
+    # R-VULN-05 BLOCKER: a scan that kills the container must cost that image only.
+    eng, runner, clock = engine(tmp, runner=KillingRunner(POSTGRES, tmp))
+    killed = False
+    try:
+        eng.tick()
+    except Killed:
+        killed = True
+    check("kill: the simulated OOM really aborted the process path", killed)
+    check("kill: inFlight is on disk BEFORE the scan starts", runner.on_disk == D4)
+    eng2, runner2, clock2 = engine(tmp, runner=FakeRunner(), clock=clock)
+    check("kill: startup clears inFlight", eng2.state["inFlight"] is None)
+    check("kill: clearing is persisted", json.loads(Path(tmp, "state.json").read_text())["inFlight"] is None)
+    eng2.publish()
+    t = targets(eng2)
+    check("kill: the target is failed", t[POSTGRES]["state"] == "failed")
+    check("kill: the reason is the killed-scan text", t[POSTGRES]["reason"] == main.KILLED_REASON)
+    eng2.tick()
+    t = targets(eng2)
+    check("kill: images after the killed one get scanned", PRIVATE in runner2.scans and t[PRIVATE]["state"] == "credential-required")
+    check("kill: the killed image is not retried on restart (no wedge)", POSTGRES not in runner2.scans)
+    check("kill: the killed target stays failed with its reason", t[POSTGRES]["reason"] == main.KILLED_REASON)
+    check("kill: inFlight is clear after a normal cycle", eng2.state["inFlight"] is None)
+    clock.advance(hours=21)
+    eng2.tick()
+    check("kill: the killed image is retried after the sit-out", runner2.scans.count(POSTGRES) == 1)
+    check("kill: a successful retry clears the failure", targets(eng2)[POSTGRES]["state"] == "scanned")
+    check("kill: a successful retry clears killedAt", eng2.state["scans"][D4]["killedAt"] is None)
+
+with tempfile.TemporaryDirectory() as tmp:
+    eng, runner, clock = engine(tmp)
+    eng.refresh_inventory()
+    eng2, runner2, clock2 = engine(tmp, runner=FakeRunner(), clock=clock)
+    eng2.publish()
+    check("mutation: a restart with no inFlight fails nothing", targets(eng2)[POSTGRES]["state"] == "pending")
+
+with tempfile.TemporaryDirectory() as tmp:
+    # A registry the scanner cannot reach is credential-required, and Trivy is never invoked.
+    eng, runner, clock = engine(tmp, private=[" GHCR.io ", ""])
+    eng.tick()
+    t = targets(eng)
+    check("private registry: reported credential-required", t[PRIVATE]["state"] == "credential-required")
+    check("private registry: the reason names the registry", "ghcr.io" in t[PRIVATE]["reason"])
+    check("private registry: Trivy is never invoked for it", PRIVATE not in runner.scans and NIGHTLY not in runner.scans)
+    check("private registry: other registries still scan", NGINX in runner.scans)
+    check("private registry: not counted as failed", report.failed_over_24h(eng.state, clock() + timedelta(hours=48)) == 0)
+with tempfile.TemporaryDirectory() as tmp:
+    eng, runner, clock = engine(tmp)
+    eng.tick()
+    check("mutation: with no private registry configured Trivy is invoked", PRIVATE in runner.scans)
+
+with tempfile.TemporaryDirectory() as tmp:
+    # A corrupt state.json degrades the run that rebuilds from it, not every run thereafter.
+    Path(tmp, "state.json").write_text("{not json")
+    eng, runner, clock = engine(tmp)
+    check("loadError: set after a corrupt state.json", eng.state["loadError"])
+    eng.tick()
+    check("loadError: the rebuilding run is degraded", eng.report["run"]["status"] == "degraded")
+    check("loadError: cleared after a full run with a good inventory", eng.state["loadError"] is None)
+    clock.advance(hours=21)
+    eng.tick()
+    check("loadError: the next full run is complete", eng.report["run"]["status"] == "complete" and eng.state["lastCompleteRunAt"])
+with tempfile.TemporaryDirectory() as tmp:
+    Path(tmp, "state.json").write_text("{not json")
+    eng, runner, clock = engine(tmp)
+    eng.client.fail = True
+    eng.tick()
+    check("mutation: loadError is kept when the inventory failed", eng.state["loadError"])
+
+with tempfile.TemporaryDirectory() as tmp:
+    # lastCompleteRunAt means a report reached the store; a refused report must not advance it.
+    eng, runner, clock = engine(tmp)
+    real_publish = report.publish
+    report.publish = lambda *_: False
+    try:
+        eng.tick()
+    finally:
+        report.publish = real_publish
+    check("a refused publish does not advance lastCompleteRunAt", eng.state["lastCompleteRunAt"] is None)
+    clock.advance(hours=21)
+    eng.tick()
+    check("mutation: a successful publish does advance it", eng.state["lastCompleteRunAt"] is not None)
+
+with tempfile.TemporaryDirectory() as tmp:
+    # One upstream string costs that string. Each of these used to freeze the whole report.
+    long_reason = "x" * 3000
+    eng, runner, clock = engine(tmp, acc="- id: CVE-2026-0010\n  reason: " + long_reason + "\n  expires: 2026-12-01\n")
+    runner.outcomes[NGINX] = {"SchemaVersion": 2, "Results": [{"Vulnerabilities": [
+        {"VulnerabilityID": "CVE-UP", "Severity": "HIGH", "PkgName": "p", "InstalledVersion": "1",
+         "PrimaryURL": "HTTPS://upper.example/x", "Title": "t\ud800"}]}]}
+    eng.tick()
+    check("hostile strings: the report still publishes", json.loads(Path(tmp, "report.json").read_text())["findings"] and report.validate(eng.report) == [])
+    up = [f for f in findings(eng, NGINX) if f["id"] == "CVE-UP"][0]
+    check("hostile strings: an uppercase scheme drops only the URL", up["url"] == "")
+    check("hostile strings: a lone surrogate is removed", "\ud800" not in up["title"])
+    check("hostile strings: an over-long acceptance is ignored, the finding stays open", [f for f in findings(eng, POSTGRES) if f["id"] == "CVE-2026-0010"][0]["status"] != "accepted")
+    check("hostile strings: an over-long acceptance degrades the run", eng.report["run"]["status"] == "degraded")
+    check("hostile strings: the degraded run is not recorded as complete", eng.state["lastCompleteRunAt"] is None)
+
+with tempfile.TemporaryDirectory() as tmp:
+    # Isolation: values already in state, whatever put them there, cannot refuse the whole report.
+    eng, runner, clock = engine(tmp)
+    eng.tick()
+    clean = report.build(eng.state, [], [], clock())
+    check("mutation: a clean state is not reported as repaired", clean["run"]["status"] == "complete" and clean == report.build(eng.state, [], [], clock()))
+    record = eng.state["scans"][A1]
+    record["findings"][0].update(url="HTTP://x", title="t" * 900, severity="weird", cvss=99, package="p" * 900)
+    eng.state["database"] = {"updatedAt": "not-a-time", "version": "1"}
+    eng.state["remediation"] = [{"image": "a/b", "state": "pr-open", "url": "HTTP://x", "updatedAt": "2026-11-09T00:00:00Z"}]
+    target = next(i for i in eng.state["inventory"] if i["id"] == NIGHTLY)
+    target["id"] = "ghcr.io/" + "o" * 3000 + ":1"
+    target["image"] = "ghcr.io/" + "o" * 3000
+    eng.state["scans"][target["id"]] = eng.state["scans"].pop(NIGHTLY)  # a by-tag unit is keyed by its id
+    built = report.build(eng.state, [], [], clock())
+    check("isolation: the repaired report is schema-valid", report.validate(built) == [])
+    check("isolation: a repair degrades the run", built["run"]["status"] == "degraded")
+    bad = [f for f in built["findings"] if f["targetId"] == NGINX and f["title"] == "t" * 512][0]
+    check("isolation: the bad finding survives with sanitised fields",
+          bad["url"] == "" and bad["severity"] == "UNKNOWN" and bad["cvss"] is None and len(bad["package"]) == 256)
+    check("isolation: every other finding is untouched", len(built["findings"]) == len(clean["findings"]))
+    check("isolation: a bad remediation loses only its url", built["remediation"][0]["url"] is None)
+    check("isolation: an unusable database is reported as unavailable", built["database"]["version"] == "unavailable")
+    long_ids = [t for t in built["targets"] if len(t["id"]) > 2000]
+    check("isolation: an over-long target stays listed", len(long_ids) == 1 and len(long_ids[0]["id"]) <= 2048)
+    check("isolation: findings follow the shortened target id",
+          any(f["targetId"] == long_ids[0]["id"] for f in built["findings"]))
+
+
+class FakeProcess:
+    returncode = 0
+
+    def __init__(self, stdout):
+        self.stdout = stdout
+
+
+class VersionRunner(trivy.Runner):
+    def __init__(self, stdout):
+        super().__init__("/nonexistent", "/nonexistent")
+        self.stdout = stdout
+
+    def _run(self, args, timeout):
+        return FakeProcess(self.stdout)
+
+
+def db_info(updated, version):
+    return VersionRunner(json.dumps({"VulnerabilityDB": {"UpdatedAt": updated, "Version": version}})).db_info()
+
+
+check("db_info rejects an unparseable timestamp", db_info("yesterday", "2") is None)
+check("db_info rejects a non-string timestamp", db_info(5, "2") is None)
+check("db_info bounds the version", len(db_info("2026-11-09T06:13:52.123456789Z", "v" * 500)["version"]) == 128)
+check("mutation: db_info keeps a good database", db_info("2026-11-09T06:13:52Z", "2") == {"updatedAt": "2026-11-09T06:13:52Z", "version": "2"})
+check("a credential failure's reason is bounded", len(trivy.classify_failure("r" * 3000, "401 unauthorized").reason) <= 1024)
+
+for url, want in (("https://a.test/x", "https://a.test/x"), ("http://a.test/", "http://a.test/"), ("HTTPS://a.test/x", ""),
+                  ("HTTP://a.test/x", ""), ("Https://a.test/x", ""), ("javascript:alert(1)", ""), ("", "")):
+    check("safe_url " + repr(url), trivy.safe_url(url) == want)
+check("every url safe_url keeps satisfies the report schema",
+      report.re.search(report.SCHEMA["properties"]["findings"]["items"]["properties"]["url"]["pattern"], trivy.safe_url("https://a.test/x")))
+
 # Recorded Trivy output through the normaliser: untrusted text.
 parsed = trivy.parse_results({"SchemaVersion": 2, "Results": [{"Vulnerabilities": [
     {"VulnerabilityID": "CVE-1", "Severity": "weird", "PkgName": "p", "InstalledVersion": "1", "Title": "x\x07y",

platform/vuln-scanner/README.md

@@ -17,17 +17,48 @@ Hand-authored platform tooling (not a catalog service), reconciled by its own no
 The page is `cve.<domain>`, internal only, gated by the per-pod oauth2-proxy sidecar to `platform-admins`.
 The route is disowned from public DNS and the tunnel 404s it; PiHole answers it on the LAN.
 
-## Operator step: add the Keycloak client (NOT done by the change that added this tree)
+## Operator action required: the Keycloak client (R-VULN-04 is NOT delivered until this is done)
 
-This change adds everything around the Keycloak client but not the client. Adding it needs a new key in
-`platform/keycloak/keycloak-realm-clients.sops.yaml`, an encrypted file an agent must not decrypt; the
-L0 check `keycloak-client-secret-reconciler` then requires the client and that key to appear in the
-same change. Until this is done the page cannot complete a login (`invalid_client`, fail-closed); the
+Until these steps are done the page cannot complete a login (`invalid_client`, fail-closed). The
 scanner, its metrics and its alerts are unaffected.
 
-In one PR (the realm import does not re-run on its own for a NEW client):
+An agent cannot do this, because every route runs through one encrypted file. The client's secret
+must be the key `vuln-scanner-client-secret` in `platform/keycloak/keycloak-realm-clients.sops.yaml`:
+the reconciler mounts only that Secret, and the L0 check `keycloak-client-secret-reconciler` rejects a
+placeholder that reads any other (a second source would be declared but never healed). That check
+also requires the client and the key to land in the same change. Adding a key needs the age private
+key. Separately, a completed realm import never re-runs, so a new client has to be created in the
+live realm once whatever git says. `*.sops.yaml` is a sensitive path, so the PR needs an approval or
+an operator-posted `/approve-override` (`infra/gitea/approval-policy.json`).
 
-1. Generate one value and set it in both encrypted files, through pipes so it is never echoed:
+**Order matters.** Create the live client first: the reconciler treats every `*-client-secret` key as
+a client that must exist, and fails every 15-minute run with `NOT FOUND` otherwise
+(`KeycloakClientSecretReconcilerFailing`).
+
+1. Create the client in the live realm. No secret is passed: the reconciler sets it from git after
+   the merge, and it reads the admin credential from the pod's own environment, so nothing is echoed.
+
+   ```bash
+   kubectl --context k3s-prod -n keycloak exec keycloak-0 -- bash -c '
+     K="/opt/keycloak/bin/kcadm.sh"; C="--config /tmp/kcadm-vuln.config"
+     $K config credentials $C --server http://localhost:8080 --realm master \
+       --user "$KC_BOOTSTRAP_ADMIN_USERNAME" --password "$KC_BOOTSTRAP_ADMIN_PASSWORD"
+     D=home.keithjackson.dev
+     ID=$($K create clients $C -r homelab -i -s clientId=vuln-scanner -s "name=CVE report (oauth2-proxy)" \
+       -s protocol=openid-connect -s enabled=true -s publicClient=false -s standardFlowEnabled=true \
+       -s directAccessGrantsEnabled=false -s "redirectUris=[\"https://cve.$D/oauth2/callback\"]" \
+       -s "webOrigins=[\"https://cve.$D\"]")
+     $K create clients/$ID/protocol-mappers/models $C -r homelab -s name=groups -s protocol=openid-connect \
+       -s protocolMapper=oidc-group-membership-mapper -s "config.\"full.path\"=false" \
+       -s "config.\"id.token.claim\"=true" -s "config.\"access.token.claim\"=true" \
+       -s "config.\"userinfo.token.claim\"=true" -s "config.\"claim.name\"=groups"
+     rm -f /tmp/kcadm-vuln.config'
+   ```
+
+   `D` is the prod `BASE_DOMAIN`; check it against the `cluster-vars` ConfigMap first.
+
+2. On a branch, generate one value and set it in both encrypted files, through pipes so it is never
+   echoed. `client-secret` in the scanner's file is replaced so that the two match:
 
    ```bash
    ./infra/devcontainer/dev.sh run bash -c '
@@ -38,7 +69,7 @@ In one PR (the realm import does not re-run on its own for a NEW client):
        "[\"stringData\"][\"client-secret\"]" "\"$cs\""'
    ```
 
-2. Add to `platform/keycloak/keycloak-realm-import.yaml`, beside the `crafty` client:
+3. Add to `platform/keycloak/keycloak-realm-import.yaml`, beside the `crafty` client:
 
    ```yaml
    # spec.placeholders
@@ -72,9 +103,11 @@ In one PR (the realm import does not re-run on its own for a NEW client):
            "claim.name": groups
    ```
 
-3. Regenerate the derived URL file: `python3 ci/render-keycloak-client-urls.py --write`. Commit all
-   three, gate, merge.
+4. Regenerate the derived URL file with `python3 ci/render-keycloak-client-urls.py --write`, and add
+   `vuln-scanner-client-secret: REPLACE-ME` to `keycloak-realm-clients.sops.yaml.example`. Commit all of
+   it in one PR, gate, merge.
 
-4. After merge, the client must also exist in the live realm (a completed realm import does not re-run,
-   per `platform/keycloak/README.md`, so create it once with `kcadm` or re-trigger the import). The
-   client-secret reconciler then converges its secret and URLs.
+5. After Flux applies it, the next reconciler run sets the client's secret and URLs from git. Prove it
+   with the L2 test `keycloak-client-secrets-match-git`, then log in at `https://cve.<domain>`. The
+   `platform-admins` restriction is the sidecar's `--allowed-group`; the `groups` mapper above is what
+   supplies the claim it reads, so a user outside that group gets a 403 from oauth2-proxy.

platform/vuln-scanner/deployment.yaml

@@ -133,6 +133,12 @@ spec:
               value: /tools/ca-certificates.crt
             - name: VULN_ACCEPTANCES
               value: /config/acceptances.yaml
+            # Registries that need a credential this scanner deliberately does not hold (R-VULN-09).
+            # Comma-separated hostnames. They report credential-required WITHOUT invoking Trivy:
+            # the egress policy excludes the LAN, so a real attempt ends in a connection error and
+            # would read as `failed`, firing VulnCoverageGap permanently. Flux substitutes the domain.
+            - name: VULN_PRIVATE_REGISTRIES
+              value: gitea.${BASE_DOMAIN}
             - name: PYTHONUNBUFFERED
               value: "1"
             - name: PYTHONDONTWRITEBYTECODE
@@ -222,7 +228,7 @@ spec:
           env:
             # client-secret MUST equal the Keycloak `vuln-scanner` client's secret (key
             # vuln-scanner-client-secret in platform/keycloak/keycloak-realm-clients.sops.yaml; see
-            # README.md, Operator step). Secrets by name only (R-IDP-04).
+            # README.md, Operator action required). Secrets by name only (R-IDP-04).
             - name: OAUTH2_PROXY_CLIENT_ID
               valueFrom:
                 secretKeyRef: { name: vuln-scanner-oauth2-proxy, key: client-id }

platform/vuln-scanner/src/acceptances.py

@@ -15,6 +15,7 @@ from datetime import date, timedelta
 from state import utcnow
 
 MAX_DAYS = 180
+MAX_REASON = 2048  # report.schema.json's limit; a longer reason would make the whole report invalid
 _KEYS = ("id", "image", "reason", "expires")
 _ENTRY = re.compile(r"^- ([a-z]+):(?: (.*))?$")
 _FIELD = re.compile(r"^  ([a-z]+):(?: (.*))?$")
@@ -91,6 +92,8 @@ def _validate(fields, today):
         problems.append("id is missing or malformed")
     if not fields.get("reason"):
         problems.append("reason is mandatory")
+    elif len(fields["reason"]) > MAX_REASON:
+        problems.append("reason is longer than " + str(MAX_REASON) + " characters")
     if "image" in fields and not fields["image"]:
         problems.append("image is empty")
     expires = fields.get("expires") or ""

platform/vuln-scanner/src/main.py

@@ -23,15 +23,21 @@ log = logging.getLogger("vuln-scanner")
 FULL_RUN_INTERVAL = timedelta(hours=20)
 DB_MAX_AGE = timedelta(hours=48)
 DB_REFRESH_AGE = timedelta(hours=24)
+KILLED_REASON = "scanner killed during scan (OOM/eviction?)"
 
 
 class Engine:
-    def __init__(self, data_dir, client, runner, acceptances_path, clock=st.utcnow, tick_seconds=900, publish_seconds=120):
+    def __init__(self, data_dir, client, runner, acceptances_path, clock=st.utcnow, tick_seconds=900, publish_seconds=120,
+                 private_registries=()):
         self.store = st.Store(data_dir)
         self.data_dir = data_dir
         self.client = client
         self.runner = runner
         self.acceptances_path = acceptances_path
+        # Registries known to need credentials the scanner deliberately does not hold (R-VULN-09) and,
+        # being on the LAN, not reachable from its egress policy: a scan would end in a connection
+        # error (`failed`) rather than a 401 (`credential-required`), so it is never attempted.
+        self.private_registries = {name.strip().lower() for name in private_registries if name.strip()}
         self.clock = clock
         self.tick_seconds = tick_seconds
         self.publish_seconds = publish_seconds
@@ -43,6 +49,18 @@ class Engine:
         self.loop_thread = None
         self._last_publish = 0.0
         self._last_inventory = None
+        self._recover_interrupted_scan()
+
+    def _recover_interrupted_scan(self):
+        """An inFlight key at startup means the last process died inside that scan."""
+        key = self.state.get("inFlight")
+        if not key:
+            return
+        log.error("the previous process was killed while scanning %s; marking it failed", key)
+        record = self.state["scans"].get(key)
+        if record is not None:
+            record["killedAt"] = fmt_ts(self.clock())
+        self._record_failure(key, "failed", KILLED_REASON)
 
     # --- views used by the HTTP servers; none of them waits for a scan --------------------------
     def report_bytes(self):
@@ -67,16 +85,19 @@ class Engine:
 
     # --- persistence -----------------------------------------------------------------------------
     def publish(self):
+        """True when the report was written; False when it was refused and the previous one stands."""
         with self.lock:
             now = self.clock()
             entries, errors = acceptances.load(self.acceptances_path, now.date())
             for problem in errors:
                 log.error("invalid acceptance ignored: %s", problem)
             built = report_mod.build(self.state, entries, errors, now)
-            if report_mod.publish(built, self.data_dir):
+            published = report_mod.publish(built, self.data_dir)
+            if published:
                 self.report = built
                 self.report_json = report_mod.serialise(built)
             self._last_publish = time.monotonic()
+            return published
 
     # --- inventory -------------------------------------------------------------------------------
     def refresh_inventory(self):
@@ -152,6 +173,14 @@ class Engine:
         return timedelta.max if moment is None else self.clock() - moment
 
     def _scan_unit(self, key, reference):
+        registry = trivy.registry_of(reference)
+        if registry in self.private_registries or registry.split(":", 1)[0] in self.private_registries:
+            self._record_failure(key, "credential-required",
+                                 "registry " + registry + " is private and the scanner holds no credential for it")
+            return
+        with self.lock:
+            self.state["inFlight"] = key
+            self.store.save(self.state)
         try:
             found = trivy.parse_results(self.runner.scan(reference))
         except trivy.ScanFailure as exc:
@@ -166,11 +195,24 @@ class Engine:
         except Exception:  # pylint: disable=broad-except
             log.exception("scan cache cleanup failed")
 
+    def _finish_in_flight(self, key):
+        """Cleared in the same write as the scan's outcome, so no crash can separate the two."""
+        if self.state.get("inFlight") == key:
+            self.state["inFlight"] = None
+
+    def _killed_recently(self, key):
+        """A unit that killed the scanner sits out full runs for a day: retrying it first on every
+        restart would OOM again before any later image, which is the wedge this exists to prevent."""
+        killed = parse_ts(self.state["scans"][key].get("killedAt"))
+        return killed is not None and self.clock() - killed < FULL_RUN_INTERVAL
+
     def _record_failure(self, key, kind, reason):
         now = fmt_ts(self.clock())
         with self.lock:
+            self._finish_in_flight(key)
             record = self.state["scans"].get(key)
             if record is None:
+                self.store.save(self.state)
                 return
             record["lastAttemptAt"] = now
             record["outcome"] = kind
@@ -181,10 +223,13 @@ class Engine:
     def _record_success(self, key, found):
         now = fmt_ts(self.clock())
         with self.lock:
+            self._finish_in_flight(key)
             record = self.state["scans"].get(key)
             if record is None:
+                self.store.save(self.state)
                 return
             record["lastAttemptAt"] = now
+            record["killedAt"] = None
             record["lastSuccessAt"] = now
             record["outcome"] = "ok"
             record["reason"] = None
@@ -215,7 +260,10 @@ class Engine:
         started = self.clock()
         log.info("full run starting")
         database_ok = self._refresh_database(always=True)
-        self._scan_queue(sorted(self._units()))
+        self._scan_queue([key for key in sorted(self._units()) if not self._killed_recently(key)])
+        if self.stop.is_set():
+            log.info("full run interrupted by shutdown; not recorded")
+            return
         info = self.state["database"]
         now = self.clock()
         degraded = not database_ok or info is None or self._db_age(info) > DB_MAX_AGE
@@ -225,8 +273,15 @@ class Engine:
         status = "degraded" if degraded else "complete"
         with self.lock:
             self.state["run"] = {"startedAt": fmt_ts(started), "finishedAt": fmt_ts(now), "status": status}
-            if not degraded:
+            self.store.save(self.state)
+            # Only a report that actually reached the store counts as a completed run. Otherwise a
+            # refused report would freeze the page while VulnScanStale stayed quiet.
+            if self.publish() and not degraded and self.report["run"]["status"] == "complete":
                 self.state["lastCompleteRunAt"] = fmt_ts(now)
+            # A corrupt state.json degrades the run that rebuilds from it, and no further: once a full
+            # run has completed against a good inventory, the state is whole again.
+            if self.state.get("loadError") and self._last_inventory is not None and not self.state.get("inventoryError"):
+                self.state["loadError"] = None
             self.store.save(self.state)
         log.info("full run finished: %s", status)
 
@@ -262,7 +317,9 @@ def main():
     tools_dir = os.environ.get("VULN_TOOLS_DIR", "/tools")
     trivy_bin = os.environ.get("TRIVY_BIN", os.path.join(tools_dir, "trivy"))
     runner = trivy.Runner(trivy_bin, cache_dir, int(os.environ.get("VULN_SCAN_TIMEOUT_SECONDS", "600")))
-    engine = Engine(data_dir, inventory.KubeClient(), runner, os.environ.get("VULN_ACCEPTANCES", "/config/acceptances.yaml"))
+    private_registries = os.environ.get("VULN_PRIVATE_REGISTRIES", "").split(",")
+    engine = Engine(data_dir, inventory.KubeClient(), runner, os.environ.get("VULN_ACCEPTANCES", "/config/acceptances.yaml"),
+                    private_registries=private_registries)
     server.start_all(engine, ("127.0.0.1", 8080), ("0.0.0.0", 8081), ("0.0.0.0", 9090))
     finished = threading.Event()
     signal.signal(signal.SIGTERM, lambda *_: finished.set())

platform/vuln-scanner/src/page.py

@@ -180,14 +180,39 @@ def _sev(severity):
     return '<span class="sev-' + e(severity) + '">' + e(severity) + "</span>"
 
 
-def _workload_lines(target):
+def _state_label(target):
+    return "target " + e(target["state"]) + (" (" + e(target["reason"]) + ")" if target.get("reason") else "")
+
+
+def _status_label(finding):
+    """One finding's own status, with its own acceptance: never another image's."""
+    label = "status " + e(finding["status"])
+    acceptance = finding.get("acceptance")
+    if acceptance:
+        label += " (" + e(acceptance["reason"]) + ", until " + e(acceptance["expires"]) + ")"
+    return label
+
+
+def _workload_lines(target, finding=None):
+    """One line per workload, each carrying the target state and, in a CVE expansion, the status."""
+    suffix = " — " + _state_label(target) + (" — " + _status_label(finding) if finding else "")
     items = []
     for workload in target["workloads"]:
         items.append("<li>" + e(workload["namespace"]) + " / " + e(workload["kind"]) + " / " + e(workload["name"])
-                     + " — containers: " + e(", ".join(workload["containers"]) or "none") + "</li>")
+                     + " — containers: " + e(", ".join(workload["containers"]) or "none") + suffix + "</li>")
     return "<ul>" + "".join(items) + "</ul>"
 
 
+def _status_summary(statuses):
+    """The row-level status. A CVE accepted on one image and open on another reads as mixed."""
+    counts = {}
+    for status in statuses:
+        counts[status] = counts.get(status, 0) + 1
+    if len(counts) == 1:
+        return e(next(iter(counts)))
+    return "mixed: " + ", ".join(e(counts[s]) + " " + e(s) for s in sorted(counts, key=STATUS_RANK.get))
+
+
 def _remediation_cell(images, remediation):
     if not images:
         return ""
@@ -250,13 +275,10 @@ def _cves(report, targets, params, remediation):
         for finding, target in row["items"]:
             if finding["status"] == "fix-available":
                 fix_images.add(target["image"])
-            lines.append("<li>" + e(target["id"]) + " [" + e(target["state"]) + "] — " + e(finding["package"]) + " "
-                         + e(finding["installedVersion"]) + " → " + e(finding["fixedVersion"] or "no fix")
-                         + _workload_lines(target) + "</li>")
-        acceptance = next((f.get("acceptance") for f, _ in row["items"] if f.get("acceptance")), None)
-        status_text = e(row["status"])
-        if acceptance:
-            status_text += " (" + e(acceptance["reason"]) + ", until " + e(acceptance["expires"]) + ")"
+            lines.append("<li>" + e(target["id"]) + " — " + _state_label(target) + " — " + _status_label(finding)
+                         + " — " + e(finding["package"]) + " " + e(finding["installedVersion"]) + " → "
+                         + e(finding["fixedVersion"] or "no fix") + _workload_lines(target, finding) + "</li>")
+        status_text = _status_summary(f["status"] for f, _ in row["items"])
         html_rows.append(
             "<tr><td>" + _link(row["url"], row["id"]) + "</td><td>" + _sev(row["severity"]) + "</td><td>"
             + e("" if row["cvss"] is None else row["cvss"]) + "</td><td>" + e(row["package"]) + "</td><td>" + e(row["fix"])
@@ -317,12 +339,13 @@ def _workloads(report, targets, params):
     html_rows = []
     for row in page_rows:
         lines = "".join(
-            "<li>" + _link(f["url"], f["id"]) + " " + _sev(f["severity"]) + " " + e(f["package"]) + " — " + e(f["status"]) + "</li>"
-            for f in sorted(row["findings"].values(), key=lambda f: (SEV_RANK[f["severity"]], f["id"])))
+            "<li>" + _link(f["url"], f["id"]) + " " + _sev(f["severity"]) + " " + e(f["package"]) + " — " + _status_label(f)
+            + " — " + e(f["targetId"]) + " — " + _state_label(targets[f["targetId"]]) + "</li>"
+            for f in sorted(row["findings"].values(), key=lambda f: (SEV_RANK[f["severity"]], f["id"], f["targetId"])))
         counts = "".join("<td>" + e(row["counts"][s]) + "</td>" for s in SEVERITIES)
         html_rows.append(
             "<tr><td>" + e(row["key"][0]) + "</td><td>" + e(row["key"][1]) + "</td><td>" + e(row["key"][2]) + "</td><td>"
-            + e(", ".join(sorted(row["containers"])) or "none") + "</td><td>" + "<br>".join(e(i) for i in sorted(row["images"])) + "</td>"
+            + e(", ".join(sorted(row["containers"])) or "none") + "</td><td>" + "<br>".join(e(i) + " — " + _state_label(targets[i]) for i in sorted(row["images"])) + "</td>"
             + counts + "<td><details><summary>" + e(len(row["findings"])) + " findings</summary><ul>" + lines + "</ul></details></td></tr>")
     head = [_th(params, "Namespace", "namespace"), _th(params, "Kind", "kind"), _th(params, "Workload", "name"), "<th>Containers</th>",
             _th(params, "Images", "images")] + [_th(params, s, s.lower()) for s in SEVERITIES] + [_th(params, "Findings", "total")]

platform/vuln-scanner/src/report.py

@@ -1,6 +1,7 @@
 """Build, validate and publish report.json, the one interface to the page and the bot."""
 import json
 import logging
+import hashlib
 import os
 import re
 from datetime import date, timedelta
@@ -85,10 +86,15 @@ def build(state, accepted, acceptance_errors, now):
     if acceptance_errors or state.get("inventoryError") or state.get("loadError") \
             or db_time is None or now - db_time > DB_STALE:
         run["status"] = "degraded"
-    return {
+    built = {
         "schemaVersion": 1, "generatedAt": fmt_ts(now), "database": database, "run": run,
         "targets": targets, "findings": findings, "remediation": list(state["remediation"]),
     }
+    repaired = _isolate(built)
+    if repaired:
+        log.error("%d report entries held a value the schema refuses; repaired or dropped them", repaired)
+        run["status"] = "degraded"
+    return built
 
 
 # --- hand validator for the JSON-Schema subset report.schema.json uses -----------------------------
@@ -175,6 +181,88 @@ def _check(schema, value, path, errors):
             _check(branch, value, path, errors)
 
 
+# --- isolation: one bad upstream value costs that value, never the whole report ---------------------
+
+def _repair(schema, value):
+    """(usable, value): value itself when valid, else the nearest valid form of it."""
+    trial = []
+    _check(schema, value, "$", trial)
+    if not trial:
+        return True, value
+    if isinstance(value, dict) and "properties" in schema:
+        fixed = {}
+        for key, sub in schema["properties"].items():
+            if key not in value:
+                continue
+            usable, item = _repair(sub, value[key])
+            if usable:
+                fixed[key] = item
+            elif key in schema.get("required", []):
+                return False, None
+        trial = []
+        _check(schema, fixed, "$", trial)
+        return not trial, fixed
+    if isinstance(value, list) and "items" in schema:
+        repaired = [_repair(schema["items"], item) for item in value]
+        return True, [item for usable, item in repaired if usable]
+    candidates = [None, "", "UNKNOWN"]
+    if isinstance(value, str) and "maxLength" in schema:
+        candidates.insert(0, value[:schema["maxLength"]])
+    for candidate in candidates:
+        trial = []
+        _check(schema, candidate, "$", trial)
+        if not trial:
+            return True, candidate
+    return False, None
+
+
+def _isolate(report):
+    """Repair (or, as a last resort, drop) each entry the schema refuses. Returns how many changed."""
+    changed = 0
+    schema = SCHEMA["properties"]
+    usable, database = _repair(schema["database"], report["database"])
+    if database is not report["database"]:
+        changed += 1
+        report["database"] = database if usable else {"updatedAt": EPOCH, "version": "unavailable"}
+    renamed, dropped, targets = {}, set(), []
+    for target in report["targets"]:
+        original = target["id"]
+        if len(original) > 2048:  # keep it unique: a truncated id may not collide with another
+            target = dict(target, id=original[:2031] + "#" + hashlib.sha256(original.encode("utf-8")).hexdigest()[:16])
+            renamed[original] = target["id"]
+        usable, fixed = _repair(schema["targets"]["items"], target)
+        if not usable:
+            changed += 1
+            dropped.add(original)
+            continue
+        if fixed is not target or original in renamed:
+            changed += 1
+        targets.append(fixed)
+    report["targets"] = targets
+    findings = []
+    for finding in report["findings"]:
+        if finding["targetId"] in dropped:
+            changed += 1
+            continue
+        if finding["targetId"] in renamed:
+            finding = dict(finding, targetId=renamed[finding["targetId"]])
+        usable, fixed = _repair(schema["findings"]["items"], finding)
+        if not usable or fixed is not finding:
+            changed += 1
+        if usable:
+            findings.append(fixed)
+    report["findings"] = findings
+    remediation = []
+    for entry in report["remediation"]:
+        usable, fixed = _repair(schema["remediation"]["items"], entry)
+        if not usable or fixed is not entry:
+            changed += 1
+        if usable:
+            remediation.append(fixed)
+    report["remediation"] = remediation
+    return changed
+
+
 def validate(report):
     """Schema and referential errors; an empty list means the report may be served."""
     errors = []

platform/vuln-scanner/src/server.py

@@ -126,7 +126,9 @@ def validate_triage(document):
         if entry["state"] not in REMEDIATION_STATES:
             return [prefix + "unknown state"]
         url = entry["url"]
-        if url is not None and (not isinstance(url, str) or len(url) > 2048 or trivy.safe_url(url) != url):
+        # safe_url is the same rule the report schema enforces, so a value accepted here cannot make
+        # the report invalid. Empty is refused: safe_url returns "" for it, and the schema does not.
+        if url is not None and (not isinstance(url, str) or not url or trivy.safe_url(url) != url):
             return [prefix + "url must be null or an http(s) URL"]
         if parse_ts(entry["updatedAt"]) is None or len(entry["updatedAt"]) > 64:
             return [prefix + "updatedAt must be RFC 3339"]

platform/vuln-scanner/src/state.py

@@ -72,6 +72,9 @@ def new_state():
         "remediation": [],
         "triageHeartbeatAt": None,
         "loadError": None,
+        # The scan key Trivy is running against right now, persisted BEFORE the scan starts: an OOM
+        # kill takes the whole container, so nothing can record the failure afterwards.
+        "inFlight": None,
     }
 
 
@@ -84,6 +87,7 @@ def new_record(now):
         "outcome": None,
         "reason": None,
         "failedSince": None,
+        "killedAt": None,
         "findings": [],
         "firstSeen": {},
     }

platform/vuln-scanner/src/trivy.py

@@ -5,6 +5,8 @@ import re
 import subprocess
 from urllib.parse import urlsplit
 
+from state import parse_ts
+
 log = logging.getLogger("vuln-scanner.trivy")
 
 SEVERITIES = ("CRITICAL", "HIGH", "MEDIUM", "LOW", "UNKNOWN")
@@ -13,7 +15,11 @@ _CREDENTIAL = re.compile(
     r"|status(?: code)?:? ?(?:401|403)\b|\b40[13] (?:unauthorized|forbidden)",
     re.IGNORECASE,
 )
-_CONTROL = re.compile(r"[\x00-\x1f\x7f]")
+# Lone surrogates too: json.loads accepts them, and they cannot be encoded as UTF-8 for the page.
+_CONTROL = re.compile(r"[\x00-\x1f\x7f\ud800-\udfff]")
+# Exact and lowercase, matching report.schema.json, whose pattern is case-sensitive: an `HTTPS://x`
+# accepted here would make the schema refuse the entire report.
+_URL_SCHEME = re.compile(r"https?://")
 
 
 class ScanFailure(Exception):
@@ -35,13 +41,18 @@ def text(value, limit=512):
 def safe_url(value):
     """value if it is a plain http(s) URL, else the empty string."""
     url = text(value, 2048)
-    if not url or url != value or re.search(r"\s", url):
+    if not url or url != value or re.search(r"\s", url) or not _URL_SCHEME.match(url):
         return ""
     try:
         parts = urlsplit(url)
     except ValueError:
         return ""
-    return url if parts.scheme.lower() in ("http", "https") and parts.netloc else ""
+    return url if parts.netloc else ""
+
+
+def registry_of(reference):
+    """The lowercase registry host (and port) of a normalised registry/repository reference."""
+    return reference.split("/", 1)[0].lower()
 
 
 def _cvss(vuln):
@@ -95,7 +106,7 @@ def parse_results(document):
 def classify_failure(registry, stderr):
     """Sort a non-zero Trivy exit into credential-required or failed. Reasons name the registry."""
     if _CREDENTIAL.search(stderr or ""):
-        return ScanFailure("credential-required", "registry " + registry + " refused anonymous access")
+        return ScanFailure("credential-required", "registry " + text(registry, 253) + " refused anonymous access")
     lines = [line.strip() for line in (stderr or "").splitlines() if line.strip()]
     return ScanFailure("failed", text(lines[-1], 400) if lines else "trivy exited with an error and no message")
 
@@ -125,9 +136,9 @@ class Runner:
             db = json.loads(result.stdout).get("VulnerabilityDB")
         except (ValueError, AttributeError):
             return None
-        if not isinstance(db, dict) or not db.get("UpdatedAt"):
-            return None
-        return {"updatedAt": db["UpdatedAt"], "version": str(db.get("Version", "unknown"))}
+        if not isinstance(db, dict) or parse_ts(db.get("UpdatedAt")) is None:
+            return None  # an unparseable timestamp is no usable database, and must not reach the report
+        return {"updatedAt": db["UpdatedAt"], "version": text(str(db.get("Version", "unknown")), 128) or "unknown"}
 
     def scan(self, reference):
         args = [
@@ -143,7 +154,7 @@ class Runner:
         if result.returncode < 0:
             raise ScanFailure("failed", "trivy killed by signal " + str(-result.returncode) + " (out of memory?)")
         if result.returncode != 0:
-            raise classify_failure(reference.split("/", 1)[0], result.stderr)
+            raise classify_failure(registry_of(reference), result.stderr)
         try:
             return json.loads(result.stdout)
         except ValueError:

renovate.json5

@@ -483,6 +483,16 @@
       "description": "ADDED 2026-10-30. MAJORS WAIT FOR A TICK ON THE DEPENDENCY DASHBOARD instead of opening a PR. They are never auto-merged anyway, and with prConcurrentLimit 5 the dry run's 10 majors would hold every PR slot open indefinitely, starving the patch/minor/digest PRs that can merge on green.",
       "matchUpdateTypes": ["major"],
       "dependencyDashboardApproval": true
+    },
+    // ADDED 2026-11-09 — EVERY IMAGE IN THE SCANNER POD, NOT ONLY TRIVY. The python image runs the
+    // process that holds the cluster-read token (rbac.yaml, R-VULN-09), and without this rule its
+    // digest and patch bumps fall through to the auto-merge rule above. LAST ON PURPOSE: Renovate
+    // applies packageRules in order and a later match overrides, so nothing above can undo it.
+    {
+      "description": "Anything under platform/vuln-scanner: the python image holds the cluster-read token. Never auto-merged, 7-day minimum release age (R-VULN-09).",
+      "matchFileNames": ["platform/vuln-scanner/**"],
+      "automerge": false,
+      "minimumReleaseAge": "7 days"
     }
   ],
   // WHAT RENOVATE STILL CANNOT FIX, so nobody mistakes a working bot for full coverage. A