@oneuptime/common 11.5.12 → 11.6.0

package/Server/Utils/Telemetry/ExceptionSanitizer.ts

@@ -0,0 +1,222 @@
+import ToolResultSerializer from "../AI/Toolbox/Serializer";
+
+/*
+ * Shared exception-text sanitization.
+ *
+ * normalizeExceptionText is the fingerprint normalizer that used to live
+ * in App/FeatureSet/Telemetry/Utils/Exception.ts — it was moved here (the
+ * App util now delegates) so server-side consumers in Common (the AI
+ * agent data API) can sanitize exception messages before they reach LLM
+ * prompts, pull-request text, and commit messages. The replacement
+ * behavior MUST stay byte-for-byte stable: fingerprints are computed from
+ * its output, and changing it regroups every existing exception.
+ */
+
+/**
+ * Normalizes a string by replacing dynamic values with placeholders.
+ * This ensures that exceptions with the same root cause but different
+ * dynamic values (like IDs, timestamps, etc.) get the same fingerprint.
+ *
+ * @param text - The text to normalize (message or stack trace)
+ * @returns The normalized text with dynamic values replaced
+ */
+export function normalizeExceptionText(text: string): string {
+  if (!text) {
+    return "";
+  }
+
+  let normalized: string = text;
+
+  // Order matters! More specific patterns should come before generic ones.
+
+  // 1. UUIDs (e.g., 550e8400-e29b-41d4-a716-446655440000)
+  normalized = normalized.replace(
+    /[0-9a-f]{8}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{4}-[0-9a-f]{12}/gi,
+    "<UUID>",
+  );
+
+  // 2. MongoDB ObjectIDs (24 hex characters)
+  normalized = normalized.replace(/\b[0-9a-f]{24}\b/gi, "<OBJECT_ID>");
+
+  /*
+   * 3. Stripe-style IDs (e.g., sub_xxx, cus_xxx, pi_xxx, ch_xxx, etc.)
+   * These have a prefix followed by underscore and alphanumeric characters
+   */
+  normalized = normalized.replace(
+    /\b(sub|cus|pi|ch|pm|card|price|prod|inv|txn|evt|req|acct|payout|ba|btok|src|tok|seti|si|cs|link|file|dp|icr|ii|il|is|isci|mbur|or|po|qt|rcpt|re|refund|sku|tax|txi|tr|us|wh)_[A-Za-z0-9]{10,32}\b/g,
+    "<STRIPE_ID>",
+  );
+
+  /*
+   * 4. Generic API/Service IDs - alphanumeric strings that look like IDs
+   * Matches patterns like: prefix_alphanumeric or just long alphanumeric strings
+   * Common in many services (AWS, GCP, etc.)
+   */
+  normalized = normalized.replace(
+    /\b[a-z]{2,10}_[A-Za-z0-9]{8,}\b/g,
+    "<SERVICE_ID>",
+  );
+
+  // 5. JWT tokens (three base64 segments separated by dots)
+  normalized = normalized.replace(
+    /\beyJ[A-Za-z0-9_-]*\.eyJ[A-Za-z0-9_-]*\.[A-Za-z0-9_-]+\b/g,
+    "<JWT>",
+  );
+
+  // 6. Base64 encoded strings (long sequences, likely tokens or encoded data)
+  normalized = normalized.replace(/\b[A-Za-z0-9+/]{40,}={0,2}\b/g, "<BASE64>");
+
+  // 7. IP addresses (IPv4)
+  normalized = normalized.replace(
+    /\b\d{1,3}\.\d{1,3}\.\d{1,3}\.\d{1,3}\b/g,
+    "<IP>",
+  );
+
+  // 8. IP addresses (IPv6) - simplified pattern
+  normalized = normalized.replace(
+    /\b(?:[0-9a-fA-F]{1,4}:){7}[0-9a-fA-F]{1,4}\b/g,
+    "<IPV6>",
+  );
+  normalized = normalized.replace(/\b::1\b/g, "<IPV6>"); // localhost IPv6
+
+  // 9. Email addresses
+  normalized = normalized.replace(
+    /\b[A-Za-z0-9._%+-]+@[A-Za-z0-9.-]+\.[A-Z|a-z]{2,}\b/g,
+    "<EMAIL>",
+  );
+
+  /*
+   * 10. URLs with dynamic paths/query params (normalize the dynamic parts)
+   * Keep the domain but normalize path segments that look like IDs
+   */
+  normalized = normalized.replace(
+    /\/[0-9a-f]{8,}(?=\/|$|\?|#|\s|'|")/gi,
+    "/<ID>",
+  );
+
+  /*
+   * 11. Timestamps in various formats
+   * ISO 8601 timestamps
+   */
+  normalized = normalized.replace(
+    /\d{4}-\d{2}-\d{2}T\d{2}:\d{2}:\d{2}(?:\.\d+)?(?:Z|[+-]\d{2}:?\d{2})?/g,
+    "<TIMESTAMP>",
+  );
+  // Unix timestamps (10 or 13 digits)
+  normalized = normalized.replace(/\b1[0-9]{9,12}\b/g, "<TIMESTAMP>");
+
+  // 12. Date formats (YYYY-MM-DD, MM/DD/YYYY, etc.)
+  normalized = normalized.replace(/\b\d{4}[-/]\d{2}[-/]\d{2}\b/g, "<DATE>");
+  normalized = normalized.replace(/\b\d{2}[-/]\d{2}[-/]\d{4}\b/g, "<DATE>");
+
+  // 13. Time formats (HH:MM:SS, HH:MM)
+  normalized = normalized.replace(/\b\d{2}:\d{2}(?::\d{2})?\b/g, "<TIME>");
+
+  // 14. Memory addresses (0x followed by hex)
+  normalized = normalized.replace(/\b0x[0-9a-fA-F]+\b/g, "<MEMORY_ADDR>");
+
+  // 15. Session IDs (common patterns) - MUST come before hex ID pattern
+  normalized = normalized.replace(
+    /\bsession[_-]?id[=:\s]*['"]?[A-Za-z0-9_-]+['"]?/gi,
+    "session_id=<SESSION>",
+  );
+
+  // 16. Request IDs (common patterns) - MUST come before hex ID pattern
+  normalized = normalized.replace(
+    /\brequest[_-]?id[=:\s]*['"]?[A-Za-z0-9_-]+['"]?/gi,
+    "request_id=<REQUEST>",
+  );
+
+  // 17. Correlation IDs - MUST come before hex ID pattern
+  normalized = normalized.replace(
+    /\bcorrelation[_-]?id[=:\s]*['"]?[A-Za-z0-9_-]+['"]?/gi,
+    "correlation_id=<CORRELATION>",
+  );
+
+  // 18. Transaction IDs - MUST come before hex ID pattern
+  normalized = normalized.replace(
+    /\btransaction[_-]?id[=:\s]*['"]?[A-Za-z0-9_-]+['"]?/gi,
+    "transaction_id=<TRANSACTION>",
+  );
+
+  // 19. Hex strings that are likely IDs (8+ chars)
+  normalized = normalized.replace(/\b[0-9a-f]{8,}\b/gi, "<HEX_ID>");
+
+  /*
+   * 20. Quoted strings containing IDs or dynamic values
+   * Match strings in single or double quotes that look like IDs
+   */
+  normalized = normalized.replace(/'[A-Za-z0-9_-]{16,}'/g, "'<ID>'");
+  normalized = normalized.replace(/"[A-Za-z0-9_-]{16,}"/g, '"<ID>"');
+
+  // 21. Port numbers in URLs or connection strings
+  normalized = normalized.replace(/:(\d{4,5})(?=\/|$|\s)/g, ":<PORT>");
+
+  /*
+   * 22. Line numbers in stack traces (keep for context, but normalize large numbers)
+   * This normalizes specific line/column references that might vary
+   */
+  normalized = normalized.replace(/:\d+:\d+\)?$/gm, ":<LINE>:<COL>)");
+
+  // 23. Process/Thread IDs
+  normalized = normalized.replace(/\bPID[:\s]*\d+\b/gi, "PID:<PID>");
+  normalized = normalized.replace(/\bTID[:\s]*\d+\b/gi, "TID:<TID>");
+
+  // 24. Numeric IDs in common patterns (id=123, id: 123, etc.)
+  normalized = normalized.replace(/\bid[=:\s]*['"]?\d+['"]?/gi, "id=<ID>");
+
+  // 25. Large numbers that are likely IDs (more than 6 digits)
+  normalized = normalized.replace(/\b\d{7,}\b/g, "<NUMBER>");
+
+  return normalized;
+}
+
+/**
+ * Sanitize an exception message for surfaces that leave the platform:
+ * LLM prompts, pull-request titles/bodies, and commit messages.
+ *
+ * Normalization replaces the dynamic tokens (UUIDs, emails, IPs, IDs,
+ * timestamps...) that carry user data, then the secret redactor sweeps
+ * whatever is left (bearer tokens, API keys, cards...). The message
+ * keeps its structure — "invalid input syntax for type uuid: <HEX_ID>"
+ * is still perfectly actionable for a fix.
+ */
+export function sanitizeExceptionMessage(message: string): string {
+  if (!message) {
+    return "";
+  }
+  return ToolResultSerializer.redact(normalizeExceptionText(message)).text;
+}
+
+/**
+ * Sanitize a stack trace for the same surfaces.
+ *
+ * Runtimes prefix the native stack string with the exception MESSAGE
+ * ("Error: <message>" in Node, "ValueError: <message>" in Python
+ * headers, "Caused by: ..." in Java) — so a redact-only pass would leak
+ * the very identifiers sanitizeExceptionMessage strips. Frame lines are
+ * indented ("    at fn (file:42:7)", '  File "x.py", line 3', "\tat ...")
+ * and must keep exact file:line references for the code agent, so the
+ * split is: unindented header/message lines get full normalization,
+ * indented frame lines are left intact, and the secret redactor sweeps
+ * everything at the end.
+ */
+// Frame lines are indented; header/message lines are not.
+const INDENTED_FRAME_LINE_REGEX: RegExp = /^\s/;
+
+export function sanitizeStackTrace(stackTrace: string): string {
+  if (!stackTrace) {
+    return "";
+  }
+
+  const normalized: string = stackTrace
+    .split("\n")
+    .map((line: string): string => {
+      return INDENTED_FRAME_LINE_REGEX.test(line)
+        ? line
+        : normalizeExceptionText(line);
+    })
+    .join("\n");
+
+  return ToolResultSerializer.redact(normalized).text;
+}