fix(otel,metrics): reject bare string classification_values, whitespace-padded entries, and reserved '_other' bucket (#113)

packages/typescript/src/metrics.ts

@@ -7,9 +7,10 @@ export class OTelMetricEmitter {
   readonly #actionCount: Counter; readonly #actionDuration: Histogram; readonly #dataFlowCount: Counter;
   readonly #tokenCount: Counter; readonly #cost: Counter; readonly #classifications: ReadonlySet<string>;
   constructor(meter: Meter, options: {classificationValues?: Iterable<string>} = {}) {
+    if (typeof options.classificationValues === "string") throw new Error("classificationValues must be an iterable of strings, not a string");
     this.#classifications = new Set(options.classificationValues ?? []);
     if (this.#classifications.size > 64) throw new Error("classificationValues cannot contain more than 64 entries");
-    for (const value of this.#classifications) if (!value || value.length > 128) throw new Error("classificationValues must be non-empty and at most 128 characters");
+    for (const value of this.#classifications) if (typeof value !== "string" || !value || value.length > 128 || value !== value.trim() || value === "_other") throw new Error("classificationValues must be non-empty and at most 128 characters");
     this.#policyCount = meter.createCounter("agentrust.policy.decisions", {unit: "{decision}", description: "Policy decisions"});
     this.#policyDuration = meter.createHistogram("agentrust.policy.evaluation.duration", {unit: "s", description: "Policy evaluation duration"});
     this.#approvalCount = meter.createCounter("agentrust.approval.events", {unit: "{event}", description: "Approval lifecycle events"});

packages/typescript/test/metrics.test.ts

@@ -25,6 +25,8 @@ test("classification dimensions are bounded by an explicit allowlist", () => {
   emitter.emit(fixture("data-flow.json"));
   assert.equal(meter.instruments.get("agentrust.data_flow.events")!.calls[0]!.attributes!["agentrust.data_flow.classification"], "_other");
   assert.throws(() => new OTelMetricEmitter(new RecordingMeter() as unknown as Meter, {classificationValues: Array.from({length: 65}, (_, i) => `class-${i}`)}), /more than 64/);
+  assert.throws(() => new OTelMetricEmitter(new RecordingMeter() as unknown as Meter, {classificationValues: "public" as unknown as Iterable<string>}), /not a string/);
+  for (const bad of [["_other"], [" public "], [123 as unknown as string]]) assert.throws(() => new OTelMetricEmitter(new RecordingMeter() as unknown as Meter, {classificationValues: bad}), /classificationValues/);
 });
 
 test("policy, approval, and action metrics expose only bounded semantic dimensions", () => {

src/agentrust_telemetry/otel.py

@@ -35,11 +35,21 @@ def __init__(
         *,
         classification_values: frozenset[str] = frozenset(),
     ) -> None:
-        if len(classification_values) > 64:
+        if isinstance(classification_values, (str, bytes)):
+            raise ValueError("classification_values must be an iterable of strings, not a string")
+        normalized = frozenset(classification_values)
+        if len(normalized) > 64:
             raise ValueError("classification_values cannot contain more than 64 entries")
-        if any(not value or len(value) > 128 for value in classification_values):
+        if any(
+            not isinstance(value, str)
+            or not value
+            or len(value) > 128
+            or value != value.strip()
+            or value == "_other"
+            for value in normalized
+        ):
             raise ValueError("classification_values must be non-empty and at most 128 characters")
-        self._classification_values = classification_values
+        self._classification_values = normalized
         self._policy_count = meter.create_counter(
             "agentrust.policy.decisions", unit="{decision}", description="Policy decisions"
         )

tests/test_otel_projectors.py

@@ -142,6 +142,10 @@ def create_histogram(self, *args, **kwargs):
             OTelMetricEmitter(
                 Meter(), classification_values=frozenset(f"class-{i}" for i in range(65))
             )
+        for bad_vals in ("public", b"public", frozenset({"_other"}), frozenset({" public "}), frozenset({123})):
+            with self.subTest(bad_vals=bad_vals):
+                with self.assertRaises(ValueError):
+                    OTelMetricEmitter(Meter(), classification_values=bad_vals)
 
     def test_action_metrics_exclude_action_and_agent_identifiers(self):
         from opentelemetry.sdk.metrics import MeterProvider