Report privately through GitHub's Report a vulnerability button on the Security tab of this repository. That opens a private advisory visible only to the maintainers.

Please do not open a public issue for a vulnerability.

A useful report includes what you did, what happened, and the smallest input that reproduces it. If the input is a real Loop document, describe its shape rather than attaching it — a .loop file carries its authors' names and email addresses (see Handling Loop files below).

Expect an acknowledgement within a week. This is a small tool maintained alongside other work, so please allow reasonable time before disclosing.

This tool parses untrusted binary input, which is the main thing worth attacking. In scope:

• memory exhaustion or unbounded allocation from a crafted container;
• a panic, hang or unbounded loop reached from a crafted container;
• reading outside a buffer, or any unsoundness;
• attribution or other personal data appearing in output that is documented as not carrying it.

Out of scope:

• inaccurate reconstruction of a document. The format is undocumented and parsed from observation; wrong output is a bug, but not a vulnerability. Open an issue.

• Bounded decompression. Each gzip member is capped, and so is the total across a file (--max-member-bytes, --max-total-bytes). A member that hits the cap is truncated and reported.
• Bounded scanning. The fragment scanner limits how far it looks back for an enclosing object, how many candidates it tries, how large an object it will match, and how many payloads it accepts from one member.
• Bounded nesting. The framing reader refuses input nested deeper than 256 levels rather than exhausting the stack.
• No guessing on unknown input. An unrecognised framing tag stops the decode and falls back to content heuristics, rather than assuming a field width and misreading everything after it.
• No network access. The tool makes no outbound connections. It needs no Microsoft 365, OneDrive, SharePoint or Graph access.

The code has not been fuzzed. That is the obvious next step for anyone wanting to harden it, and parse in src/discovery/envelope.rs plus match_object in src/discovery/fragment_scanner.rs are the two entry points worth targeting.

A .loop container is a personnel record as well as a document. Nearly every paragraph carries its author's display name, email address, directory object id and an edit timestamp; a single page of meeting notes can hold well over a hundred such records.

Consequences worth knowing:

• inspect --dump-payloads writes raw payloads and does include all of it. Treat the output directory as you would the source file.
• Never attach a real .loop file to an issue or a pull request.
• Never commit one. The repository ignores the fixture directories used for local testing.