human-agent-trust-reviewer

What problem does it solve?

This Skill helps evaluate whether an agent approval layer can be manipulated through consent fatigue, polished but misleading justifications, batching tricks, urgency pressure, or automation bias, so human approval remains a real control instead of a ceremonial click.

Core Features & Use Cases

• Approval-flow adversarial review: Checks what approvers actually see, what one approval authorizes, and whether the request is based on system-verified facts or the agent’s own story.
• Fatigue and trust calibration analysis: Uses telemetry such as request volume, approval rate, and decision time to detect rubber-stamping and recommends recalibrating gate friction to match blast radius.
• Bundling and deception resistance: Looks for dangerous actions hidden inside innocuous batches, repeated re-asks after denial, and urgency framing that tries to lower scrutiny.
• Use case: Review a production change approval flow where operators approve almost everything in seconds and the agent supplies its own one-line summary before execution.

Quick Start

Ask the skill to review your agent approval flow for consent fatigue, self-report gaps, bundling, and other trust exploits, and return ranked findings with concrete fixes.