Skip to content

macOS Inspector 1.4.4

Choose a tag to compare

@bacescuandrei bacescuandrei released this 04 Oct 14:36
· 6 commits to main since this release

macOS Inspector 1.4.4

This release makes process follow-up more precise, shows the local rule context behind a result, and keeps delayed dashboard responses from mixing different scans.

Process follow-up

After a recorded process action, use Process action history > Recheck process outcome. The follow-up is a new read-only Live Triage snapshot, not another termination request.

The result distinguishes the original identity still present, a reused PID, a zombie, the original PID no longer observed, and insufficient evidence. It now also shows when the same executable path and numeric owner appear under another PID. That observation may describe an existing instance or a restart; it proves neither. A matching path does not establish identical file contents or malicious activity.

Follow-ups require matching recorded host context and valid, timezone-aware action, scan-window, and process-evidence timestamps. Prior or overlapping scans, duplicate PIDs, incomplete inventories, and missing historical context remain unverifiable. New actions record the local hostname. Older action records remain visible but cannot supply missing identity or host evidence.

Normal Live Triage process results use Observed, which the follow-up now accepts correctly. Process inventory completeness is recorded explicitly. Unparsed rows, duplicate PIDs, missing numeric owners, command failures, and collection bounds prevent absence inferences without discarding the processes that were observed.

Local rules and provenance

When IOC or YARA was selected, About this scan > Local rules and provenance shows the recorded rule snapshot separately from result availability. The same context is included in the standalone HTML report and investigation summary.

Expand Recorded rule details and provenance for the context SHA-256, fingerprinted file count, and bounded IOC pack metadata. Pack versions, update labels, and source hostnames are declared by the pack, not independently authenticated. Source URL credentials, query strings, fragments, and full source paths are not copied into the derived summary. Opening the panel does not contact a source.

A stable fingerprint does not mean the rules are reliable or that checks ran. Disabled YARA can have a stable fingerprint and still say Not run. Filtering can omit metadata or results; missing context remains explicit rather than implying a clean assessment.

Consistent report selection

Report evidence, decision support, and export links load as one selected snapshot. Delayed responses from an earlier selection cannot overwrite a newer scan. Decision support with a different scan ID is rejected while the recorded report remains available.

Process response is blocked while a report is loading. Investigation saves capture the originating scan and do not replace another scan's guidance after navigation. Candidates that are no longer displayed are removed from the dashboard's process-action map. Server-side process authorization and identity checks remain independent.

Verification

Local verification passed 121 Python tests, 68 browser tests across Chromium and WebKit, and all 26 synthetic scenarios. Browser checks include delayed JSON and decision-support responses, report identity mismatches, pending-load process controls, investigation saves during navigation, narrow layouts, escaped provenance, and targeted accessibility checks.

The portable package was rebuilt deterministically, checked after extraction, and tested with local read-only collection. Git integrity and privacy checks passed. GitHub CI repeats Linux/macOS tests on Python 3.10 and 3.13, both browser engines, and portable packaging. Publication follows a successful CI run and a byte-for-byte comparison of its artifact with the local build.

Boundaries

Collectors remain read-only. Process signals still require explicit confirmation, a recent authorized snapshot, current-user ownership, and live identity revalidation. No automatic response, reputation database, telemetry, new online provider, or evidence upload was added.

Hostnames and owner-controlled logs are not authenticated identities or tamper-proof evidence. Process collection and rule capture are not atomic. An absent PID does not establish containment or resolve an investigation. Online intelligence remains opt-in.

The dashboard assumes a trusted local session. Browser-origin and loopback checks are not per-client authentication; another local program or account can call the API. Close the dashboard when finished and do not run it as a shared service on a host with untrusted local users. This release clarifies that boundary; it does not add client authentication.

The action log retains one previous file and shows up to 50 recent actions. Source findings remain unchanged. The 26 scenarios are regression fixtures, not an independent detection benchmark. WebKit automation does not establish coverage for every Safari version or screen reader.

Python 3.10 or newer is still required. The portable release is not Apple-signed or notarized. No runtime installation, privilege elevation, or change to macOS security settings is automatic.

Install and artifact

Download the ZIP and SHA256SUMS from this release, compare the complete SHA-256, extract the archive, and double-click macOS Inspector.command. If a compatible runtime is missing, use the included docs/START_HERE.html.

Filename: macos-inspector-1.4.4-macos.zip
SHA-256: a9f8b6734c318bab25d40b95b09ad026c790324d213e6927b59a5eb2fc435618

See the README, Usage guide, Validation guide, and Changelog.