Escalate critical AI R&D decisions to human operators for approval.
npx skills add https://github.com/TenureAI/PhD-Zero --skill human-checkpoint
Please help me install this Agent Skill. Skill: human-checkpoint Source: https://github.com/TenureAI/PhD-Zero/tree/main/.agents/skills/human-checkpoint Command: npx skills add https://github.com/TenureAI/PhD-Zero --skill human-checkpoint
This Skill ensures that critical AI research and development decisions, especially those involving significant risks or requiring human oversight, are escalated appropriately to a human for approval.
full-auto, moderate, detailed).Use the human checkpoint skill to request a decision on a potentially destructive operation.
High-intent search queries and answers about installing and using this skill.
Human-in-the-loop escalation pauses AI execution to request operator approval for critical decisions. It assesses safety risks and presents structured options with tradeoffs, preventing irreversible actions like large-scale file deletion without explicit human consent.
Implement human approval by triggering a checkpoint that pauses execution and presents a decision request. Operators choose to allow once, allow for the run, or disallow, ensuring high-impact destructive mutations receive explicit human oversight.
You need a human checkpoint when an AI workflow encounters hard blockers, high-resource approvals, or major safety risks like secret leakage. It ensures irreversible impact actions are escalated to human operators before execution proceeds.
Yes, you can adjust escalation triggers using full-auto, moderate, and detailed modes. Mode-aware escalation adapts interaction policies based on active mode, changing how and when human approval is required.
Automated risk assessment cannot execute irreversible actions itself; it is limited to flagging major safety risks and hard blockers. It requires a human operator to review the presented tradeoffs and explicitly approve, disallow, or allow the action.
jshsakuraIdentify responsible-AI risks and assess user impact in AI features.
m-ahmad-officialScan pending approval directories and list items awaiting human authorization.
argenCapture and manage project decisions in a schema-validated structured format.
FVossebeldEnforce layered guardrails for prompts, tool usage, and outputs in Microsoft-native agent contexts.