This project is not distributed as a versioned package. Security fixes are made on the main branch, which is the only supported version.

Please do not open a public issue for a suspected vulnerability. Instead, use GitHub's private vulnerability reporting and include:

• a description of the vulnerability and its impact;
• the affected files, configuration, and environment;
• the smallest reproducible example you can provide; and
• any suggested remediation, if you have one.

Do not include credentials, personal data, or live exploit targets in the report. Please allow time to investigate and coordinate a fix before public disclosure.

Useful reports include vulnerabilities in the repository's code or default configuration, such as:

• bypasses of public-destination validation, DNS pinning, or redirect checks;
• ways to introduce a briefing URL that is absent from the fetched corpus;
• parser or resource-exhaustion flaws that violate the documented input bounds;
• credential exposure from the optional evaluator; and
• prompt-injection paths that cross a boundary the project claims to enforce.

The following are documented limitations rather than vulnerabilities by themselves:

• inaccurate ranking or summaries produced by a model;
• prompt injection that changes selection or prose without escaping the corpus URL allowlist;
• unsafe tools granted to the generating agent contrary to the deployment precondition in briefing-prompt.md;
• malicious, inaccurate, unavailable, or rate-limited third-party feeds; and
• denial of service against the upstream public sources.

The README describes the trust boundaries and guarantees in more detail. Reports showing that an implementation does not meet one of those stated guarantees are in scope.