@base-org/account@2.4.0 security report

The serializeError function includes the full error message (serialized.message) and error code as query parameters in a docUrl pointing to an external domain (docs.cloud.coinbase.com). If this error is later logged, sent to analytics, or shared, sensitive information from the error message could leak to third parties. While likely benign (intended for documentation links), it constitutes an outbound data flow containing potentially sensitive error details.

Setting the serialized error message into a URL query parameter (docUrl.searchParams.set('message', serialized.message)) can expose arbitrary error content (which may include secrets, paths, or user data) in logs, referrers, or browser history. This is a privacy/security concern despite no direct malicious intent.

All three functions transmit subscription metadata (correlationId, amount, permissionHash, errorMessage, testnet flag, periods) to an external analytics endpoint via logEvent. This is expected telemetry behavior but constitutes outbound data flow that should be reviewed for what fields are permitted.

logSubscriptionError forwards data.errorMessage verbatim to logEvent. Error messages frequently contain sensitive values (keys, hashes, tokens, URLs) and could exfiltrate them to the analytics endpoint without sanitization.

A hardcoded Amplitude API key ('c66737ad47ec354ced777935b0af822e') is embedded directly in the source. Even public telemetry keys should not be hardcoded, as it enables unauthorized telemetry injection or abuse.

The code trusts any object named window.ClientAnalytics and calls logEvent/identify on it. This creates a hook point where external scripts (browser extensions, compromised dependencies, or injected code) could intercept analytics data or trigger side effects. No allowlist or integrity check exists.

The bundle uses a custom webpack-style module loader with runtime-computed requires (n(2), n(353), n(762), etc.) and dynamic export assignment, plus UMD-style fallback attaching ClientAnalytics to the global object. This is normal bundler output but makes auditing harder and could mask malicious payloads in minified form.

The code stores configuration (including metadata, preferences, and paymaster URLs) in a persistent store (store.config.set + store.persist.rehydrate). This persists user-supplied configuration across sessions, which is expected but worth noting for the privacy-sensitive nature of crypto SDK data.

getInjectedProvider() may return a browser-injected provider (e.g., window.ethereum). This allows third-party code to be used as the provider, which is standard for wallet SDKs but means the actual signing/request handling is outsourced to whatever provider is injected.

The function accesses window.top, which may throw a SecurityError when the script runs inside a cross-origin iframe. This can cause runtime exceptions and could be used to probe or infer frame hierarchy relationships. It is not inherently malicious but is a fragile and potentially risky pattern.

Error messages leak internal details such as the sender wallet address and details of all USDC transfers found (Found ${usdcTransfers.length} USDC transfer(s)..., transfer details including recipient addresses). While not exploitable on its own, verbose error content could aid reconnaissance in multi-tenant or sensitive contexts.

Telemetry calls (logPaymentStarted/logPaymentCompleted/logPaymentError) pass the payment amount, address, and error messages. While these appear to be local event loggers, if the underlying telemetry module forwards events to a remote server, it constitutes data exfiltration of financial details. The implementation of '../../core/telemetry/events/payment.js' is not visible.

Uses a non-standard 'wallet_sign' method through provider.request() with mutableData fields, requesting signatures for spend permissions. While this appears to be a legitimate subscription flow using EIP-712 typed data, the custom method and placeholder address replacement mechanism could be abused if the wallet implementation is compromised or if the signing flow is manipulated.

Imports from multiple internal modules (telemetry, spend-permission utilities, SDK manager) whose implementations are not visible in this file. These could contain malicious code. The telemetry module in particular could perform network requests.

The function relies on getPermissionStatus (likely an RPC/offchain read) to decide whether to include approveWithSignature, then constructs spend and transfer calls locally. There is no cryptographic binding preventing a race between status check and execution, but this is inherent to the approval flow rather than a direct malicious pattern.

The optional recipient is used as a raw address argument to encodeFunctionData for ERC20 transfer without checksum validation, zero-address check, or confirmation that it relates to the intended permission flow. Malformed or malicious addresses supplied by callers could misroute funds.

Keys are generated with extractable:false, which is a positive security control. No exfiltration, obfuscation, dynamic execution, network calls, or process spawning was detected. The only external interactions are imports from trusted libraries (ox, viem) and local storage operations.

Signing payloads are bound to the hardcoded origin 'https://keys.coinbase.com'. This is intentional WebAuthn origin binding, but it means signatures are only valid for that origin and could be leveraged in phishing scenarios if the user is misled about the relying party.

getItem, setItem, and removeItem do not handle errors from idb-keyval. In a key-management context, a failed write or delete could leave stale or inconsistent key material, but no error is surfaced to the caller.

The scope and name arguments are passed directly to createStore without validation, allowing callers to choose arbitrary IndexedDB database/object store names. This could be abused to interfere with other components or to persist data in unexpected locations.

The class stores cryptographic keys (private and public) in a local store. While this is expected for key management, private keys stored in plaintext within the application store could be exposed if the store is compromised or if storage is not properly secured. This is a common pattern in wallet/SDK implementations and does not itself indicate malicious behavior.

The code fetches data from external RPC URLs (e.g., CB_WALLET_RPC_URL and this.chain.rpcUrl) via fetchRPCRequest. While this is expected for a wallet signer (communicating with blockchain RPC nodes), the constant CB_WALLET_RPC_URL is imported from core/constants.js and its value is not visible here; if it pointed to an attacker-controlled endpoint, it could exfiltrate data. This warrants inspection of that constant.

The code reads window.location.origin to build a data suffix for attribution. This is a benign use for dapp attribution but demonstrates access to browser environment data.

No eval, new Function, or dynamic import from untrusted input is present. Module imports are static.

No child_process or shell command execution is present.

No access to .npmrc, ~/.ssh, ~/.aws, or other credential files is present.

Functions logDialogActionClicked, logDialogDismissed, and logDialogShown are imported from telemetry modules and invoked when displaying dialogs. This suggests user interaction data (dialog context, action) is being recorded/transmitted. While common in legitimate analytics, it constitutes data collection from the user environment.

assertFetchPermissionsRequest and fillMissingParamsForFetchPermissions construct spend permission requests using a spender address from the store. This is legitimate wallet functionality but represents sensitive approval-granting logic that should be carefully validated.

The exported 'keys.set(key, value)' accepts arbitrary string keys and values and stores them directly into the persisted store without any validation, allow-list, or encryption. A caller (or any code with access to the imported module) could write arbitrary data into localStorage under the 'base-acc-sdk.store' namespace, and the same accessor could be used to read back whatever was stored.

The openPopup function uses window.open with a computed popupId and URL that receives appended query parameters. While common for OAuth/wallet flows, an attacker controlling the URL argument could potentially craft a malicious popup if the caller does not validate the URL. This is a design risk rather than an active malicious pattern.

The appendAppInfoQueryParams function appends origin (window.location.origin) and Cross-Origin-Opener-Policy (coop) values to the popup URL. While this is used to pass SDK metadata (name, version) for compatibility/diagnostics, it also leaks the dapp's origin and COOP state to the URL target. If the URL is later logged, cached, or redirected, this information could be exposed. However, the URL is used with window.open in the same context and the values are expected for wallet SDK operations, so this is a low-severity information disclosure rather than exfiltration.

The code calls logDialogShown and logDialogActionClicked to send telemetry events about dialog interactions. These functions are imported from the package's own telemetry module, which may transmit user interaction data to a remote endpoint. While standard for SDK analytics, without inspecting the telemetry implementation, this could be a privacy concern.