ResumeShield is a defensive tool. It exists to detect and neutralize hidden instructions in candidate documents and to measure how well a screening pipeline resists them.
The benchmark contains injection payloads. They are generic, publicly documented examples of the kind catalogued under OWASP LLM01:2025 (for instance, "ignore all previous instructions"). They are included for the same reason an intentionally vulnerable test application is included in a scanner project: a defense that cannot be measured cannot be trusted. The project does not publish novel evasion techniques, and pull requests whose primary contribution is a stronger attack with no corresponding defense or measurement will not be merged.
Do not use this project to manipulate a hiring system you do not own. Submitting a resume containing hidden instructions to a real employer is, at minimum, a misrepresentation and may be unlawful depending on jurisdiction.
All resumes and payloads shipped here are synthetic, generated from a seeded random generator. No real candidate data is present. If you evaluate ResumeShield against real documents, treat them as personal data under the applicable regulation and do not commit them to a repository.
Please open a private GitHub Security Advisory rather than a public issue.