blakstrip redacts sensitive documents entirely in the browser. Security is the product, so we take reports seriously.

Please do not open a public issue for security problems. Instead, report privately through GitHub's private vulnerability reporting ("Report a vulnerability" under the repository's Security tab).

Include, where possible:

• what leaks or fails (residual text, visible ink under a box, surviving metadata, network egress),
• a minimal PDF or steps that reproduce it (redact synthetic data, never real personal data),
• the browser and version.

We aim to acknowledge a report within a few days and to ship a fix or mitigation before any public disclosure.

In scope: anything that breaks the core promise, such as content that survives a redaction, hidden data that export fails to strip, or any network request leaving the page (the production build ships a strict connect-src 'none' Content-Security-Policy).

Out of scope: issues that require a compromised device, a malicious browser extension, or a modified build.

blakstrip is honest about what it proves. Redacted pages are rasterized: rendered to a flat image with the black boxes burned into the pixels, so the text and vectors underneath are gone, not merely hidden. Untouched pages are copied verbatim and keep selectable text.

Verify-on-export re-opens the output and checks three things: (1) no redacted term is still recoverable as text, (2) no structural leak vector (metadata, XMP, annotations, forms, JavaScript, attachments, EXIF/GPS, optional-content layers) is still present in it, and (3) the redacted regions actually read black in the output raster (a pixel-coverage check, so a box that visually under-covers a glyph is caught before download rather than certified "clean").

Verify does not block the download. It shows you exactly what survived and labels the button "Download anyway" when anything did, because an imperfect file is sometimes still the file you want. Anything listed in red means the output still leaks: cancel, fix the redaction, and export again.

Verify cannot reason about content it was never told to redact. Redact everything sensitive, review the audit panel on load, and read the verify dialog before downloading.

• EXIF/GPS inside a photo is detected, not stripped. Metadata attached to an embedded image is removed, but camera and GPS data baked into the JPEG's own bytes would require re-encoding the image. Both the audit and the verify dialog flag it. To remove it, draw a redaction box anywhere on the page holding that image: the page is then rasterized and re-encoded, and the EXIF goes with it.
• Optional-content (layer) PDFs are not fully cleaned. On a page copied verbatim, layer content stays in the file, and the exported file loses the layer visibility settings, so a layer that was hidden may become visible. blakstrip detects this and refuses a clean verdict; to actually remove it, redact something on every page that uses layers so those pages get flattened.
• Encrypted PDFs are refused. pdf-lib cannot decrypt, so a password- or permission-protected file has to be re-saved without protection first.
• Untouched pages keep selectable text. Only pages you redact are rasterized. That is what keeps the rest of the document searchable, and it is why the verify dialog lists what is still readable.