Offline archive safety¶

openmed.interop.archive_safety classifies archive-member metadata before a file-redaction workflow extracts anything. It is deliberately local-only: the module does not open archives, read member contents, follow links, or make network calls.

Decisions¶

inspect_archive_members returns an ArchiveSafetyReport with one of three decisions:

• allow: the supplied metadata is within policy and has no structural findings.
• quarantine: resource limits were exceeded. A separate review must approve the archive before extraction.
• reject: traversal, links, duplicate normalized paths, malformed metadata, or another structural safety finding was detected. The archive must not be extracted.

Diagnostics contain reason codes and counts only. Member paths and link targets are not copied into reports, serialized output, or exception messages. Report reason iteration and member kind normalization are bounded before caller-controlled values are processed.

from openmed.interop.archive_safety import (
    ArchiveDecision,
    ArchiveSafetyPolicy,
    inspect_archive_members,
)

policy = ArchiveSafetyPolicy(
    max_entries=1_000,
    max_total_uncompressed_bytes=64 * 1024 * 1024,
    max_expansion_ratio=50,
)
report = inspect_archive_members(
    [
        {
            "path": "records/summary.txt",
            "compressed_size": 120,
            "uncompressed_size": 480,
            "kind": "file",
        }
    ],
    policy,
)

assert report.decision is ArchiveDecision.ALLOW

The evaluator treats both / and \\ as archive separators, rejects absolute and parent-traversal paths, detects normalized duplicate paths, rejects symbolic and hard links, and quarantines entry-count, size, and expansion-ratio limit breaches. The default limits are intentionally bounded; callers may tighten them for a specific workflow. Size metadata is restricted to unsigned 64-bit values, and report aggregates are bounded, so hostile Python integers cannot make diagnostics or later serialization unbounded.