fix: improve Sentry issue grouping to eliminate duplicate issues

src/lib/error-reporting.ts

@@ -21,14 +21,17 @@ import * as Sentry from "@sentry/node-core/light";
 import {
   ApiError,
   AuthError,
+  CliError,
   ContextError,
   DeviceFlowError,
+  HostScopeError,
   OutputError,
   ResolutionError,
   SeerError,
   TimeoutError,
   UpgradeError,
   ValidationError,
+  WizardError,
 } from "./errors.js";
 
 // ---------------------------------------------------------------------------
@@ -101,6 +104,43 @@ function recordSilencedError(error: unknown, reason: SilenceReason): void {
 // Grouping tags
 // ---------------------------------------------------------------------------
 
+/** Endpoint normalization patterns — compiled once at module scope. */
+const ENDPOINT_PATTERNS: [RegExp, string][] = [
+  [/\/organizations\/[^/]+/, "/organizations/{org}"],
+  [/\/projects\/[^/]+\/[^/]+/, "/projects/{org}/{project}"],
+  [/\/issues\/[^/]+/, "/issues/{id}"],
+  [/\/events\/[^/]+/, "/events/{id}"],
+  [/\/groups\/[^/]+/, "/groups/{id}"],
+  [/\/releases\/[^/]+/, "/releases/{version}"],
+  [/\/teams\/[^/]+\/[^/]+/, "/teams/{org}/{team}"],
+  [/\/dashboards\/[^/]+/, "/dashboards/{id}"],
+  [/\/customers\/[^/]+/, "/customers/{org}"],
+];
+
+/**
+ * Strip remaining bare numeric segments (e.g. /12345/) but preserve
+ * the API version prefix /0/ which is always the second segment.
+ */
+const BARE_NUMERIC_SEGMENT_RE = /(?<=\/api\/0\/.*)\/\d+(?=\/|$)/g;
+
+/**
+ * Normalize an API endpoint path by parameterizing variable segments.
+ *
+ * Replaces org slugs, project slugs, issue IDs, event IDs, and other
+ * entity identifiers with placeholders so that server-side fingerprint
+ * rules can sub-group `ApiError` by endpoint shape rather than exact path.
+ *
+ * `"/api/0/projects/my-org/my-project/events/abc123/"` →
+ * `"/api/0/projects/{org}/{project}/events/{id}/"`
+ */
+export function normalizeEndpoint(endpoint: string): string {
+  let result = endpoint;
+  for (const [pattern, replacement] of ENDPOINT_PATTERNS) {
+    result = result.replace(pattern, replacement);
+  }
+  return result.replace(BARE_NUMERIC_SEGMENT_RE, "/{id}");
+}
+
 /**
  * Strip quoted substrings, numeric/hex IDs, and org/project paths from a
  * resource string to produce a stable "kind" for grouping.
@@ -111,14 +151,24 @@ function recordSilencedError(error: unknown, reason: SilenceReason): void {
  * `"not found in neurio/installer-app"` → `"not found"`
  */
 export function extractResourceKind(resource: string): string {
-  return resource
-    .replace(/'[^']*'/g, "")
-    .replace(/"[^"]*"/g, "")
-    .replace(/\b[0-9a-f]{16,32}\b/gi, "")
-    .replace(/\bin\s+[\w-]+\/[\w-]+/g, "")
-    .replace(/\b\d+\b/g, "")
-    .replace(/\s+/g, " ")
-    .trim();
+  return (
+    resource
+      .replace(/'[^']*'/g, "")
+      .replace(/"[^"]*"/g, "")
+      .replace(/\b[0-9a-f]{16,32}\b/gi, "")
+      .replace(/\bin\s+[\w-]+(?:\/[\w-]+)*/g, "")
+      // Strip hyphenated slugs after known entity names (e.g., "Organization my-company").
+      // Requires at least one hyphen to avoid stripping English words ("Project not found").
+      // Safe for current callers: resource values with slugs use quotes (stripped above),
+      // and headline values don't start with entity names.
+      .replace(
+        /\b(Organization|Dashboard|Dashboards|Project|Team)\s+[\w][\w-]*-[\w-]*/gi,
+        "$1"
+      )
+      .replace(/\b\d+\b/g, "")
+      .replace(/\s+/g, " ")
+      .trim()
+  );
 }
 
 /**
@@ -140,6 +190,64 @@ export function extractMessagePrefix(message: string, maxWords = 3): string {
     .join(" ");
 }
 
+/**
+ * Derive a stable `cli_error.kind` grouping key from an error instance.
+ *
+ * Returns `undefined` when the error is not a recognized CLI error class
+ * (the caller should still set `cli_error.class` for basic grouping).
+ */
+function deriveErrorKind(error: Error): string | undefined {
+  if (error instanceof ContextError) {
+    return error.resource;
+  }
+  if (error instanceof ResolutionError) {
+    return (
+      extractResourceKind(error.resource) +
+      " " +
+      extractResourceKind(error.headline)
+    );
+  }
+  // Fall back to the first few words of the message when no field is set
+  // (e.g. validateHexId throws with no `field`, so using field would
+  // collapse every unfielded ValidationError into one group).
+  if (error instanceof ValidationError) {
+    return error.field ?? extractMessagePrefix(error.message);
+  }
+  if (error instanceof ApiError) {
+    return String(error.status);
+  }
+  if (error instanceof SeerError) {
+    return error.reason;
+  }
+  if (error instanceof AuthError) {
+    return error.reason;
+  }
+  if (error instanceof UpgradeError) {
+    return error.reason;
+  }
+  if (error instanceof DeviceFlowError) {
+    return error.code;
+  }
+  if (error instanceof TimeoutError) {
+    return "timeout";
+  }
+  if (error instanceof HostScopeError) {
+    return "host_scope";
+  }
+  if (error instanceof WizardError) {
+    return "wizard";
+  }
+  // Catch-all for bare CliError — must be checked AFTER all subclasses
+  // because instanceof matches the entire prototype chain.
+  // ConfigError and OutputError intentionally fall through here:
+  // ConfigError has no structured field beyond message; OutputError is
+  // silenced by classifySilenced() before reaching deriveErrorKind().
+  if (error instanceof CliError) {
+    return extractMessagePrefix(error.message, 4);
+  }
+  return;
+}
+
 /**
  * Set `cli_error.*` tags on a Sentry scope for an error that will be
  * captured. These tags are matched by server-side fingerprint rules to
@@ -149,6 +257,7 @@ export function extractMessagePrefix(message: string, maxWords = 3): string {
  * - `cli_error.class` — error class name (e.g. `"ContextError"`)
  * - `cli_error.kind`  — stable grouping key derived from structured fields
  * - `cli_error.api_status` — HTTP status (ApiError only)
+ * - `cli_error.api_endpoint` — normalized API path (ApiError only)
  */
 function setGroupingTags(scope: Sentry.Scope, error: unknown): void {
   if (!(error instanceof Error)) {
@@ -157,36 +266,16 @@ function setGroupingTags(scope: Sentry.Scope, error: unknown): void {
 
   scope.setTag("cli_error.class", error.name);
 
-  if (error instanceof ContextError) {
-    scope.setTag("cli_error.kind", error.resource);
-  } else if (error instanceof ResolutionError) {
-    scope.setTag(
-      "cli_error.kind",
-      extractResourceKind(error.resource) +
-        " " +
-        extractResourceKind(error.headline)
-    );
-  } else if (error instanceof ValidationError) {
-    // Fall back to the first few words of the message when no field is set
-    // (e.g. validateHexId throws with no `field`, so using field would
-    // collapse every unfielded ValidationError into one group).
-    scope.setTag(
-      "cli_error.kind",
-      error.field ?? extractMessagePrefix(error.message)
-    );
-  } else if (error instanceof ApiError) {
+  const kind = deriveErrorKind(error);
+  if (kind !== undefined) {
+    scope.setTag("cli_error.kind", kind);
+  }
+
+  if (error instanceof ApiError) {
     scope.setTag("cli_error.api_status", String(error.status));
-    scope.setTag("cli_error.kind", String(error.status));
-  } else if (error instanceof SeerError) {
-    scope.setTag("cli_error.kind", error.reason);
-  } else if (error instanceof AuthError) {
-    scope.setTag("cli_error.kind", error.reason);
-  } else if (error instanceof UpgradeError) {
-    scope.setTag("cli_error.kind", error.reason);
-  } else if (error instanceof DeviceFlowError) {
-    scope.setTag("cli_error.kind", error.code);
-  } else if (error instanceof TimeoutError) {
-    scope.setTag("cli_error.kind", "timeout");
+    if (error.endpoint) {
+      scope.setTag("cli_error.api_endpoint", normalizeEndpoint(error.endpoint));
+    }
   }
 }
 
@@ -273,5 +362,12 @@ export function enrichEventWithGroupingTags(
   event.tags = event.tags ?? {};
   event.tags["cli_error.class"] = exc.type;
 
+  // Set kind from exception message prefix so server-side rules can group
+  // non-CliError exceptions (TypeError, Error, WizardCancelledError, etc.)
+  // that bypass reportCliError (uncaught exceptions, unhandled rejections).
+  if (exc.value) {
+    event.tags["cli_error.kind"] = extractMessagePrefix(exc.value, 4);
+  }
+
   return event;
 }

src/lib/telemetry.ts

@@ -313,6 +313,35 @@ export function isEpipeError(event: Sentry.ErrorEvent): boolean {
   return false;
 }
 
+/**
+ * Detect EBADF (bad file descriptor) errors in Sentry events.
+ *
+ * These occur when the init wizard's stdin reopen (`stdin-reopen.ts`) queues
+ * reads on a destroyed file descriptor. Same class of OS-level noise as EPIPE
+ * — not actionable, just different fd numbers producing duplicate issues.
+ *
+ * @internal Exported for testing
+ */
+export function isEbadfError(event: Sentry.ErrorEvent): boolean {
+  const exceptions = event.exception?.values;
+  if (exceptions) {
+    for (const ex of exceptions) {
+      if (ex.value?.includes("EBADF")) {
+        return true;
+      }
+    }
+  }
+
+  const systemError = event.contexts?.node_system_error as
+    | { code?: string }
+    | undefined;
+  if (systemError?.code === "EBADF") {
+    return true;
+  }
+
+  return false;
+}
+
 /**
  * Check if an error is a user-caused (401–499) API error.
  *
@@ -612,6 +641,13 @@ export function initSentry(
         return null;
       }
 
+      // EBADF errors come from the init wizard's stdin reopen queuing reads
+      // on a destroyed fd. Same class of OS-level noise as EPIPE — different
+      // fd numbers just produce duplicate issues. Not actionable — drop them.
+      if (isEbadfError(event)) {
+        return null;
+      }
+
       // Normalize relative frame paths to absolute. Bun's compiled binaries
       // with sourcemap: "linked" produce relative paths like "dist-bin/bin.js"
       // in Error.stack. Sentry's symbolicator only matches absolute paths

test/lib/error-reporting.property.test.ts

@@ -25,6 +25,9 @@ const slugArb = array(
   .map((chars) => chars.join(""))
   .filter((s) => !(s.startsWith("-") || s.endsWith("-")) && s.length > 0);
 
+/** Slug that contains at least one hyphen (matches the entity-name strip regex). */
+const hyphenatedSlugArb = slugArb.filter((s) => s.includes("-"));
+
 /** 32-character lowercase hex id (trace/event/log id). */
 const hexIdArb = stringMatching(/^[0-9a-f]{32}$/).filter(
   (s) => s.length === 32
@@ -130,4 +133,45 @@ describe("extractResourceKind — property tests", () => {
       { numRuns: DEFAULT_NUM_RUNS }
     );
   });
+
+  test("bare slug after 'in' (no slash) is stripped for any slug", () => {
+    fcAssert(
+      property(slugArb, slugArb, (a, b) => {
+        expect(extractResourceKind(`not found in ${a}`)).toBe(
+          extractResourceKind(`not found in ${b}`)
+        );
+        expect(extractResourceKind(`not found in ${a}`)).toBe("not found");
+      }),
+      { numRuns: DEFAULT_NUM_RUNS }
+    );
+  });
+
+  test("hyphenated slug after entity name is stripped for any slug", () => {
+    fcAssert(
+      property(hyphenatedSlugArb, hyphenatedSlugArb, (a, b) => {
+        expect(extractResourceKind(`Organization ${a}`)).toBe(
+          extractResourceKind(`Organization ${b}`)
+        );
+        expect(extractResourceKind(`Organization ${a}`)).toBe("Organization");
+      }),
+      { numRuns: DEFAULT_NUM_RUNS }
+    );
+  });
+
+  test("Dashboard with numeric ID and slug produces same kind", () => {
+    fcAssert(
+      property(
+        numericIdArb,
+        slugArb,
+        numericIdArb,
+        slugArb,
+        (n1, s1, n2, s2) => {
+          expect(extractResourceKind(`Dashboard ${n1} in ${s1}`)).toBe(
+            extractResourceKind(`Dashboard ${n2} in ${s2}`)
+          );
+        }
+      ),
+      { numRuns: DEFAULT_NUM_RUNS }
+    );
+  });
 });