Security audit
Security checks for vulnerabilities and agentic risk
This privacy tool needs review because its optional report exposes sensitive information that the sanitized document removes.
Do not share --report output: it can contain complete personal details and secret excerpts. Seek a version that removes raw samples and accurately declares file access. Limit access to explicitly selected inputs and manually review sanitized results because detection uses limited patterns. Host enforcement of the declared restrictions remains unverified.
The documented file-input workflow conflicts with filesystem: false, while the optional report exposes original sensitive values rather than only safe audit metadata. Including that report with an externally shared sanitized document can disclose complete phone numbers, short email addresses, and secret prefixes, directly undermining this skill's privacy purpose.
Referenced artifact was not completely inspected
With --report, the script prints up to 12 original characters from every sensitive match into the same stdout stream as the sanitized document. This reveals complete 11-digit phone numbers and short email addresses, as well as partial secrets, making output capture or external sharing unsafe.
Without declared permissions the skill's intent is opaque and cannot be validated.
The --file branch reads a caller-supplied local path despite the explicit filesystem: false declaration, with no application-level path restriction. If the host relies on that declaration rather than enforcing a sandbox, the script can access files readable by its process; this does not demonstrate a bypass of an enforced filesystem denial.