ShadowShield is a security tool, so please route reports correctly:
If you've found a prompt-injection / jailbreak / exfiltration technique that ShadowShield fails to catch, that is not a vulnerability in ShadowShield β it's a gap in coverage the whole community benefits from fixing in the open. Please:
- Open a public issue using the "Attack pattern / bypass" template.
- Include a minimal reproduction string and the expected category.
- Even better β open a PR adding a signature + a regression test.
Examples: a way to silently disable a detector, a ReDoS in a pattern that hangs the scanner, a path that leaks a secret into the audit log despite redaction, or a bypass of the fail-closed guarantees. Please do not open a public issue. Report privately via GitHub Security Advisories:
Repository Security tab β Report a vulnerability (Private Vulnerability Reporting). This opens a private channel visible only to you and the maintainers.
Include a description, reproduction steps, affected versions, and impact. We aim to acknowledge within 72 hours and to ship a fix or mitigation within 30 days for confirmed high-severity issues, coordinating disclosure with you.
| Version | Supported |
|---|---|
| 0.6.x | β |
| < 0.6 | β |
ShadowShield is a strong layered filter, not a guarantee. It is one control in a defense-in-depth strategy (least-privilege tools, human-in-the-loop on high-impact actions, output validation). Reports that amount to "a sufficiently novel prompt got through" are coverage gaps (public), not framework vulnerabilities (private) β though we want to hear about both.