RedactKit - AI Privacy Scrubber

Scan your data before sending it to AI. Detect and redact PII, secrets, and sensitive info. Reversible, local, zero network calls.

This skill appears to do exactly what it claims: regex-based, reversible local redaction. Important cautions before you install/use it: - Mapping files contain original sensitive values in plaintext (LIMITATIONS.md documents this). Treat mapping JSON like secrets: never commit them to version control, store them on encrypted filesystems, and restrict filesystem permissions (chmod 600). Consider encrypting mapping files (GPG, filesystem encryption) if they will be stored or transferred. - Regex-based detection has false positives and false negatives. Test on non-sensitive samples first, use report mode, and add exclusion/custom patterns as needed. Neither the tool nor its patterns provide semantic understanding. - Custom patterns accept arbitrary regex. Malformed or exponential-time regexes can cause crashes or slowdowns (ReDoS). Validate custom regexes and avoid pathological patterns for untrusted inputs. - The tool loads entire files into memory and may be slow or OOM on very large files; follow README guidance to split large files. - Although the code contains no network calls, the README references external preprocessing (OCR, pdftotext) as optional steps; those external tools can introduce network or additional risks—only run trusted preprocessing tools. If you need higher assurance for compliance (e.g., HIPAA, legal redaction), use a certified tool or additional safeguards—this is a useful ad-hoc/local utility but not a compliance guarantee.

Type: OpenClaw Skill Name: redact-kit Version: 1.0.0 RedactKit is a legitimate privacy tool designed to redact PII and secrets locally before data is shared with AI services. The codebase (redact_kit.py, redact_patterns.py, redact_restore.py) relies exclusively on the Python standard library, using regex for detection and local JSON files for mapping storage. No evidence of data exfiltration, network calls, or malicious execution was found, and the documentation (SKILL.md) contains no prompt injection attempts or instructions to bypass security controls.