IBM Bob Premium Package for Z

AI Security Review

Evidence context: threat category none; evidence quality strong.

What the extension does and why the access fits its purpose

This is IBM’s officially published mainframe development extension, used by nearly 30,000 developers on OpenVSX. It parses COBOL and other mainframe languages, generates docs, and visualizes diagrams. The package ships a large bundle in extension/dist/ that includes a language server (gql2tmi-server.js), a wiki rendering engine, and Mermaid diagramming. These are standard components for a mainframe IDE tool.

The three network findings match the bundle’s job. NET-FETCH-extension/dist/extension.js-1 and NET-FETCH-extension/dist/gql2tmi-server.js-1 are fetch calls inside the language server and extension core. Language servers routinely fetch schema updates or check for back-end health. NET-FETCH-extension/dist/packages/wiki/dist/vendor/mermaid.js-1 sits inside an embedded Mermaid library. That fire is a JavaScript diagram renderer making calls to load fonts or renderer assets. None of these endpoints point to unexpected external infrastructure.

No permissions are declared and the extension requests no host permissions. That alone sharply limits what any code inside the bundle can reach. An extension that wanted to exfiltrate source or credentials would need broader network surfaces than an unadorned fetch inside a chart library.

Credential access: nothing targets real secrets

The findings contain zero secret detections. No hardcoded keys, .env references, or credential-store access patterns appear. The code-smell category logged 1677 matches from bundled JavaScript, and every one is a low-severity pattern like postinstall_* rules firing on minified library code. These rules match generic Node.js constructs (fs, process.env, fetch) that exist in virtually every non-trivial npm bundle. The extension itself does not touch SSH keys, cloud credentials, or workspace secret files.

The IoC noise masks no real threat

1167 IoC extractions dominate the finding count. The endpoint list exposes what happened: fragments like 0ny.sa, 2sg.sc, 8ego.at, and aesni.inc are not real domains. They are property-access chains, hex identifiers, and assembly macro names extracted from minified JavaScript, Mermaid’s parser tables, and bundled COBOL grammar data. The string abstractcustomcobolanalysisrule.java is literally a Java class name. These are classifier artefacts, not network beacons. The XIOC extractor grabbing them from a 3MB+ bundle is expected noise.

32 dependency findings register bundled third-party code. The extension ships with the Mermaid diagramming library, wiki rendering, and a language server. All are standard. None are flagged by malware signatures. The zero malware and zero obfuscation findings across the entire scan confirm that no deliberate concealment or malicious payload exists.

Strongest counterargument

2879 total findings sounds alarming at a glance, and all-medium severity (1194) from IoC volume could make a casual reviewer think something is wrong. But the severity inflation comes from raw count alone. There are zero critical, zero high, and zero malware-signature findings. Every single medium finding is an IoC artifact or a code-smell match in a third-party bundle file. IBM’s extension is doing mainframe development work with the same JavaScript patterns every other VS Code language extension uses.

This is a well-known false-positive profile: a large, minified, multi-library bundle scanned by a system that treats any string that looks even vaguely like a domain as an indicator. The evidence quality is strong because the pattern is clear and reproducible across every bundled library file in extension/dist/.

Key Reasons

• Zero malware signatures across 2879 findings: the count is driven entirely by IoC extraction from minified library bundles
• All three network findings match standard bundled-library behaviour (language server fetch, Mermaid renderer fetch)
• No secrets detected, no credential-access patterns, and no declared host permissions
• Findings cluster in known false-positive drivers: XIOC domain extraction from hex strings, YARA code-smell rules on generic Node.js patterns
• Published by IBM with 29,380 installs; extension function (mainframe language support, wiki, Mermaid) matches all bundled code

False Positive Considerations

• XIOC domain extraction from minified JS/parser tables yielding 1167 false IoCs (e.g. '0ny.sa', 'aesni.inc', 'abstractcustomcobolanalysisrule.java')
• YARA code-smell rules matching generic Node.js patterns in dist/ libraries (1677 low-severity matches)
• Bundled third-party code (Mermaid, wiki renderer, language server) amplifying finding counts multiplicatively

Reviewed 2026-09-27; recommended action: suppress false positive; model confidence 92%.