Please do not open a public issue for anything that could expose investigation data, a user's browsing history, or a secret.

Until a dedicated private channel is configured, use GitHub's private vulnerability reporting for this repository.

Please include the affected version or commit, reproduction steps, the impact, and a suggested mitigation if you have one.

filegrail reads local records that are, by design, sensitive. The output can contain every URL a file came from and every command that touched it. Issues worth reporting:

• A secret surviving --redact. The redaction is biased towards precision, so a false negative is a real finding. Include the shape of the credential.
• Reading or writing outside the scanned directory, or outside the browser profiles the tool declares it reads. A crafted file must not be able to make the scan touch an arbitrary path.
• Modification of anything inspected. The tool copies a browser profile before opening it and never writes to a scanned file. A path that breaks that is a bug of the first order.
• A crafted file causing unbounded memory use or a hang. Every parser here reads untrusted input and carries explicit bounds; a way past them counts.
• A metadata claim rendered as more certain than it is — most importantly a C2PA manifest presented without its "signature not verified" note, since the tool does not validate the certificate chain.

Changes to these need extra review:

• redact.py, and every path that renders a URL or a command;
• the parsers under sources/embedded/, sources/c2pa.py and cbor.py, all of which read untrusted bytes;
• sources/browser.py, which copies and opens the user's browser profiles;
• archive handling in sources/archives.py;
• path handling in scan.py.