SuperAPI AI Directory

frida-data-extraction

Security & auditing

View repository

Updated 10-16 01:09

About this skill

Frida Data Extraction

Use this skill when the goal is to observe runtime data rather than bypass a check.

Rules

  • Only extract data from systems the user is authorized to test.
  • Avoid printing secrets into shared logs unless the user explicitly needs them.
  • Prefer structured logs with redaction for tokens, cookies, passwords, and private keys.
  • Copy pointer and retval data inside callbacks; Frida may recycle callback objects.

Android Boundaries

Java.perform(() => {
  const Base64 = Java.use("android.util.Base64");
  Base64.encodeToString.overload("[B", "int").implementation = function (bytes, flags) {
    const out = this.encodeToString(bytes, flags);
    console.log("Base64.encodeToString ->", out);
    return out;
  };
});

High-signal targets: request builders, JSON serializers, SharedPreferences, SQLite APIs, keystore wrappers, crypto APIs, WebView bridges, file I/O.

iOS Boundaries

High-signal targets: NSURLRequest, NSURLSession, NSJSONSerialization, Keychain wrappers, NSUserDefaults, SQLite, CommonCrypto, file APIs, pasteboard.

Native Buffers

function dump(ptr, len) {
  if (ptr.isNull() || len <= 0) return;
  console.log(hexdump(ptr, { length: Math.min(len, 256), ansi: false }));
}

Use both hex and text only when the data is valid text. Binary crypto material often is not.

Verification

  • Logs are tied to a user action or request.
  • The source and encoding of each value is clear.
  • Sensitive values are redacted unless needed.
  • The hook does not change the target behavior unless explicitly requested.

References