Safe to install, safe to delegate
Every product is a plain-text Markdown + JSON file kit. It governs and documents; it never executes code, holds credentials, or takes an irreversible action on its own.
Non-destructive guarantees
New-folder install
Installs into its own buyer-chosen folder. Never overwrites existing files.
Preview before apply
Every change is previewable before it is applied. No silent, irreversible actions.
Human-approval gates
Spend, publish, delete, and credential-adjacent steps require named human approval.
Rollback included
Delete the installed folder and nothing else is touched. A rollback guide ships in every kit.
How a risky action is handled
Deny-by-default: dangerous tool calls are tiered, gated, and audited before anything runs. The T4 human-approval floor for spend, publish, delete, and credential-adjacent actions cannot be lowered.
What products never do
Execute code, or take destructive action without preview and approval.
Request or hold credentials, API keys, seed phrases, or wallet secrets.
Auto-publish, auto-post, or auto-purchase on the operator's behalf.
Defensive by design
Security products (e.g. prompt-injection defense) are strictly defensive, with regression suites.
Written contracts & verification — run alongside platform-native permissions, not instead of them.
Binding never-store lists cover credentials, keys, and sensitive identifiers.