IBM Functional Testing

AI Security Review

Evidence context: threat category none; evidence quality moderate.

ibm-functionaltest is IBM's generator for Galasa functional tests, published on OpenVSX to about 34 users. Every finding that carries critical severity is one rule: OBFUSCATION-NATIVE_BINARY_ADDON, applied eleven times to extension/prebuilds/keyring.darwin-arm64.node, keyring.darwin-x64.node, keyring.linux-x64-gnu.node, keyring.win32-x64-msvc.node and the other platform variants. Those are compiled prebuilt binaries shipped by the keyring npm module, which talks to the operating system credential store so the user does not need a compiler to install the extension. A .node file is machine code. The rule fires because string extraction fails on compiled code, which is the normal state of any native addon.

Filesystem and process access lines up with the stated job. The package declares no editor permissions and no host permissions, and nothing lands in the network, manifest-analysis or secret categories. A test generator writes Java files into the workspace and runs Gradle or Maven against them, and the evidence shows nothing wider than that.

The credential question deserves attention because keyring wraps the OS keychain, and a keychain wrapper is what a credential stealer would ship. Nothing here reads .env files, .git/config or SSH keys: the secret category is empty and no finding names those paths. The prebuilds are the library for putting an IBM Cloud API token in macOS Keychain, Windows Credential Manager or libsecret instead of a plaintext file.

The network endpoint list is mostly extractor noise. Entries like al.bo, ae.es, at.ch, 0ny.sa, csslineclass.info and data.rel.ro are substrings pulled out of minified JavaScript and Java test templates rather than real hosts. The one genuine destination is api.dataplatform.cloud.ibm.com, IBM's Cloud Pak for Data API, where a Galasa test run reports its results.

The strongest counterargument is that static analysis cannot read a native addon, so an unaudited binary in the package is a real blind spot, and keyring prebuilds are exactly the artifact a malicious extension would hide behind. That argument is sound about method and empty about evidence: across 467 findings there are zero malware signatures, zero reads of developer secrets, and no endpoint other than IBM's own API. The eleven critical labels describe the file format, not the behavior inside it.

Bundled dependency code accounts for the remaining volume. The 103 code-smell matches and 11 dependency findings sit in shipped libraries and template files, and those rules match ordinary JavaScript idioms in build output.

Key Reasons

• All 11 critical findings are OBFUSCATION-NATIVE_BINARY_ADDON on extension/prebuilds/keyring.*.node, which detects compiled machine code rather than hidden behavior
• No malware signatures, no secret-category findings, and no path-based findings for .env, .git/config or SSH keys
• The only real endpoint in the list is api.dataplatform.cloud.ibm.com, IBM's own Cloud Pak for Data API, consistent with Galasa test reporting
• Network endpoint entries such as al.bo, ae.es, at.ch and csslineclass.info are substrings of minified JS and Java templates, not contacts
• The extension is published by IBM on OpenVSX with a stated purpose that matches the file writes and process execution present

False Positive Considerations

• OBFUSCATION-NATIVE_BINARY_ADDON fires on compiled .node prebuilds in extension/prebuilds/ that cannot be string-scanned
• IoC extractor reads substrings of minified JavaScript and Java test templates as domains (al.bo, at.ch, csslineclass.info)
• 103 code-smell matches on bundled dependency JavaScript and templates
• 342 IoC entries dominated by hex fragments and CDN/infra noise rather than real destinations

Reviewed 2026-09-30; recommended action: suppress false positive; model confidence 85%.