Fixing an Approval Gate Nobody Reads

A team's coding-agent harness prompts for confirmation on every shell command. Telemetry from the last month shows: a median session has 32 command approvals, the median time-to-approve is 1.4 seconds, and 99.1% of prompts are approved. In a recent incident, an engineer approved a git push --force to a shared branch in that same 1.4 seconds, alongside the 31 pytest and ruff approvals around it.

1. Explain, using the telemetry, why this gate is not functioning as a control — and be precise about the mechanism, not just "alert fatigue."
2. Redesign the gate. State what you would change and what the telemetry should look like afterwards.
3. A teammate proposes the opposite fix: "stop prompting entirely and just log everything for review." Assess that proposal.