AI Agent Lockout Backup Kit

A vendor-independent continuity kit for backing up, rotating, and recovering information used by AI agents—without storing secrets in the agent or in GitHub.

Core principle: If an AI agent becomes unavailable, the business owner must still have independent access to every credential, configuration, document, and recovery procedure needed to continue operating.

What this project is

This repository provides:

• A credential and access inventory template
• Business-continuity and fallback planning templates
• Onboarding, offboarding, and exposure-response checklists
• A static, offline-first worksheet site for recording metadata only
• JSON schemas for provider, FAQ, and test-result content
• Content-verification rules that separate reports from verified documentation
• A clearly labeled, unverified muse-safety-protocol-proposal.md policy proposal
• A professional incident report documenting the failed Meta account-export workaround
• A consolidated, redacted docs/muse1-incident-report.md with timeline and recovery actions
• A polished Support page with optional third-party subscription and donation links
• Automated checks for navigation, data loading, and unsafe credential fields

What this project is not

• It is not official support for Muse or any other AI provider.
• It is not a password manager, secret vault, or credential-recovery service.
• It does not prove that any AI agent is safe to hold credentials.
• It does not automatically run prompts against live accounts.

Never put passwords, API keys, tokens, payment details, private customer data, or production credentials in this repository or in the website worksheets.

Quick start

1. Copy templates/credential-inventory.csv to a private, encrypted location.

2. Fill in service names, owners, recovery URLs, rotation instructions, and backup locations—not secret values.

3. Store actual secrets in a dedicated password manager or encrypted vault with at least one human-held recovery path.

4. Complete templates/business-continuity-plan.md and test the fallback plan.

5. Open site/index.html locally, or serve the site/ directory with any static server.

6. Run the checks:

   node tests/data-validation.test.js
   node tests/navigation.test.js
   node tests/worksheets.test.js
   node tests/accessibility.test.js
   node tests/support.test.js

Backup model

Use three separate layers:

1. Inventory: what exists, who owns it, where it is used, and how it is rotated.
2. Secret storage: actual values kept in a dedicated encrypted vault, outside this repository and outside a single agent.
3. Recovery evidence: redacted procedures, provider links, backup locations, and last-tested dates.

Backups are for continuity, not proof that previously shared secrets remain safe. If an agent or chat may have displayed a credential, revoke and regenerate it through the provider that issued it.

Repository layout

ai-agent-lockout-backup/
├── docs/                 Guidance and verification rules
├── templates/            Redacted, fill-in continuity documents
├── schemas/              JSON schemas for site data
├── site/                 Static worksheet and reference website
├── tests/                Dependency-free Node checks
└── .github/workflows/    GitHub Pages deployment and validation

Privacy model of the website

The worksheet stores only non-secret metadata in the browser's localStorage. It has no server, analytics, login, or network submission. Users can export or delete local worksheet data. Service-worker caching excludes worksheet state. Treat browser storage as sensitive nonetheless: use a private device and delete data when finished.

Content reliability

Every provider-specific instruction should have a source URL and a last_verified date. Label claims as user-reported, verified, or unknown. Do not turn an anecdote about a lockout into a universal platform claim.

The file muse-safety-protocol-proposal.md is preserved as a user-reported policy proposal. Its incident details and platform-specific claims are not independently verified by this repository.

License

MIT. See LICENSE.