Please do not file public GitHub issues for suspected vulnerabilities. The project prefers coordinated disclosure:

1. Open a private GitHub security advisory on this repository, or
2. Contact the maintainers privately with a minimal reproduction and an impact assessment.

A valid report includes:

• The vulnerable versions.
• A proof-of-concept (code, request, payload, target URL, etc.).
• Your assessment of impact (confidentiality / integrity / availability) and any mitigations you have already considered.

• Acknowledgement within 5 business days of receipt.
• Triage and remediation plan within 10 business days, proportional to severity (CVSS v3.1).
• Fix released as a patch version (2.0.X) out of the regular cadence for confirmed high- or critical-severity issues. Lower-severity issues ride the next scheduled release.
• Advisory published on the repository with the affected version range, mitigation instructions, and credit to the reporter (unless the reporter prefers anonymity).

In scope:

• All code under nostrax/ in this repository.
• The published nostrax distributions on PyPI.
• The CI workflow configuration at .github/workflows/.

Out of scope:

• Vulnerabilities in direct or transitive dependencies. Please report those upstream to the respective projects (aiohttp, lxml, beautifulsoup4, defusedxml, regex, packaging, tomli).
• Bugs in callers that use the library incorrectly (e.g., running the crawler without the shipped SafeResolver against an untrusted target and then complaining about SSRF).
• Social-engineering attacks on the maintainers.

nostrax ships several protections on by default. Understanding what they do - and where they stop - is part of using it safely against untrusted input.

Built-in protections:

• SSRF / DNS-rebinding: validate_target_url rejects targets that resolve to loopback, private, link-local, reserved, multicast, unspecified, or cloud-metadata addresses, and SafeResolver re-applies that classifier at every DNS resolution the crawl performs, closing the TTL=0 rebinding window.
• XXE / XML bombs: sitemap parsing goes through defusedxml with an additional DOCTYPE/ENTITY string reject.
• ReDoS: --pattern / --exclude run through the regex package with a per-URL match timeout.
• Path traversal: file and cache writes are confined to the current working directory via a fully-resolved, case-normalised containment check.
• Resource limits: response, robots.txt, and sitemap sizes are capped; the frontier queue is bounded; the read cap applies to decompressed bytes, so a gzip bomb is caught after inflation.
• Credential hygiene: userinfo is stripped from emitted URLs and redacted from logs; redirects are disabled on every request.

Known limitations (by design):

• A proxy voids SSRF protection. When --proxy is set, the proxy, not nostrax, resolves the target hostname, so SafeResolver never sees the target's address and cannot reject an internal one. Do not point a proxied crawl at untrusted targets unless the proxy itself enforces an egress policy.
• robots.txt is advisory. --respect-robots is opt-in; the default is to ignore robots.txt. When enabled, an unreachable or malformed robots.txt fails open (crawling is allowed).
• Running without SafeResolver. The library-level crawl_async installs SafeResolver itself, but a caller who supplies a custom Fetcher or their own aiohttp session is responsible for their own egress safety.

Non-security bug reports belong on the public issue tracker.