Please do not open a public issue for a vulnerability. Use GitHub's private vulnerability reporting for this repository. If that is unavailable, contact the maintainer privately through the email listed on their GitHub profile.

Include the affected version, reproduction steps, impact, and suggested mitigation. You should receive an acknowledgement within five business days.

Doubt copies static skill files, reads the map JSON explicitly passed to it, and writes the requested HTML artifact. Normal validation, rendering, installation, and browser-playground flows do not call a model, fetch sources, upload data, execute map content, run a background service, or collect telemetry.

doubt verify is an explicit, opt-in exception: it retrieves the source URLs already present in the map and compares normalized visible text with each recorded excerpt. It does not upload the map or excerpt to a Doubt service. HTTP redirects are followed manually and every destination is checked; loopback, private, link-local, carrier-grade NAT, benchmark, and multicast IP ranges are blocked unless the caller passes --allow-private. Responses are limited to text-like content, five redirects, 5 MB per source, and a default ten-second timeout. DNS rebinding remains a limitation of the built-in Node HTTP stack; do not run verification on attacker-controlled URLs from a network that can reach sensitive HTTP services.

Map strings are HTML-escaped before rendering and embedded JSON escapes markup delimiters. Source URLs remain user-supplied clickable links in the output; a valid schema does not establish that a destination is safe or truthful. Review untrusted maps before opening their external links.

Installation writes only to the selected agent skill directory. Use doubt doctor to detect missing or locally modified installed copies.