Gate privileged tool calls behind admin approval tokens and optional 2FA.
npx skills add https://github.com/misunders2d/ori --skill approval-skill
Please help me install this Agent Skill. Skill: approval-skill Source: https://github.com/misunders2d/ori/tree/main/skills/approval-skill Command: npx skills add https://github.com/misunders2d/ori --skill approval-skill
Privileged actions (delete, force-push, mass-modify, send-public-message, etc.) don't execute immediately. The plugin layer AdminGatePlugin intercepts the tool call and returns an action token instead.
Approve a pending privileged action by sending "Approve ACT-XXXXXX" (and your 6-digit code if 2FA is enabled) to authorize the queued operation.
High-intent search queries and answers about installing and using this skill.
To require admin approval for privileged actions, this Skill uses an AdminGatePlugin to intercept sensitive tool calls and return an action token instead of executing immediately. Pending operations are queued for human review until explicitly approved.
Token-based approval works by intercepting sensitive tool calls and generating a unique action token. The pending action is stored in a local SQLite database until an administrator authorizes it by submitting the token and an optional 2FA code.
Yes, you can secure sensitive operations with TOTP 2FA. When configured, the approval workflow requires a six-digit TOTP code alongside the action token to authorize queued privileged actions like mass-modify or delete.
The best way to authorize a pending privileged action is to send an approval command containing the action token, such as "Approve ACT-XXXXXX". If 2FA is enabled, you must also include your six-digit TOTP code in the same message.
Yes, you need a database to queue pending admin approvals for destructive actions. The Skill uses a local SQLite database file to store intercepted privileged tool calls until they are explicitly approved or canceled by an administrator.
You should use an approval gate for privileged tool calls whenever an action involves destructive or sensitive operations like delete, force-push, or mass-modify. This ensures human oversight before execution and prevents unintended system changes.
AgentientInitialize Firebase Admin SDK server-side in Next.js 14+ apps.
Mueza-EjazManages approval workflows for payments, emails, and social posts with audit logging.
jayzalowitzWarn before executing destructive commands like rm -rf and git reset --hard.
riecktAudit Convex applications for authorization and data access boundaries.
