A private, agent-driven security audit for an individual's Mac and core security posture. It is designed for someone who is not technical: open the repository in a fresh Claude Code or Codex session, say “start the audit,” read one clear permission explanation, and approve the read-only checks.
This repository does one job. Source-code repository analysis lives in the separate private security-audit repository.
Open this repository in a fresh agent session and say:
Start my personal security audit.
The agent explains exactly what it will inspect and asks once for consent. It then runs the automated macOS collector and returns useful results before asking any questions.
The normal audit does not need screen recording, Accessibility, microphone, camera, administrator access, or Full Disk Access. Terminal-based checks are more reliable and less invasive than visually navigating System Settings.
- macOS version, pending updates, and automatic-update configuration;
- FileVault, firewall, stealth mode, Gatekeeper, System Integrity Protection, and XProtect presence;
- lock-screen delay and display-sleep timing;
- Time Machine recency and active status;
- device-management enrollment state;
- remote login, screen sharing, file sharing, and remote-management states;
- aggregate non-loopback listener counts;
- current-account administrator status, Guest User, and automatic login;
- recognized password-manager presence;
- aggregate Chromium-family browser-extension permissions without names, IDs, or history;
- SSH private-key file permissions without reading key contents or filenames.
The collector reduces raw local output to fixed security states and aggregate counts before the agent sees it. It never outputs usernames, hostnames, IP addresses, process names, app names, extension names or IDs, filesystem paths, private-key names, passwords, vault contents, messages, personal documents, browser history, or recovery codes. Safari and Firefox extension permissions are left for the final human-only review rather than being falsely marked covered.
The report leads with confirmed corrections and their exact evidence. Each correction has an ID, severity, plain-English reason, safe next step, approval requirement, rollback, and verification. Enabled sharing or listening services are marked for intent review, not labeled malicious.
At the end, a short Could not verify automatically appendix covers account and behavior protections such as password uniqueness, MFA/passkeys, recovery-code storage, phishing habits, and backup restore testing. The audit never withholds the useful machine report while waiting for a questionnaire.
python3 scripts/validate.py
python3 skills/audit-personal-security/scripts/audit_macos.pyUse --no-network to skip Apple's update-availability query. Use --format json for the fixed-schema machine-readable result.
Read TRUST.md. It explains how to inspect the collector, pin a reviewed commit, validate the repository, watch the runtime transcript, and have a separate person or fresh agent independently review what the toolkit can do.
The current automated collector supports macOS. Windows and Linux are explicitly unsupported rather than receiving a generic or misleading audit.
No license has been selected while the repository remains private.