Repository Data Exposure Review

Audit repository exposure without transmitting, reproducing, changing, or deleting the data under review. Treat automation as candidate discovery, not proof that data is safe or anonymous.

Preserve the review boundary

• Remain read-only. Do not sanitize files, rotate credentials, rewrite Git history, delete artifacts, contact third parties, or save a report unless separately authorized.
• Keep repository content, detected values, private paths, manifests, history, and reports local. Do not submit them to websites, hosted scanners, models, registries, or other external services.
• Never print, quote, hash, or include surrounding text for a detected value. Report only the repository-relative location, category, confidence, exposure state, and remediation direction.
• Record git status --short before and after the review. Stop and report any unexpected tracked-file mutation caused by a command.
• Use only repository-provided or already-installed read-only tools. Do not install detectors, dependencies, models, OCR packages, or Git history-rewrite utilities.
• Do not claim legal compliance, regulatory certification, penetration testing, irreversible anonymization, or a guarantee that no personal data exists.

Read the focused references

• Read data-classification-and-signals.md before classifying personal/private data, public attribution, synthetic examples, re-identification risk, or migration artifacts.
• Read scope-history-and-response.md before selecting Git scope, interpreting coverage gaps, reporting committed exposure, or recommending remediation.

Establish scope and repository truth