Mirrored verbatim from utopia-models/knowledge _rules/when-tyler-is-talking-stop-executing.md by the allowlisted publish pipeline. Edit it there, not here.

name: when-tyler-is-talking-stop-executing
cases: [2026-07-01-agent-briefing-mode-pacing, 2026-07-01-cloud-brand-gen-cadence, 2026-07-01-cloud-reject-means-stop-not-pace, 2026-07-14-stop-verifying-aggregate-all-masters]
status: active
mechanism: prose — the trigger needs judgement a hook cannot observe
audience: agent
last-recurrence: 2026-07-14

When Tyler is talking, absorb. A tool call mid-briefing is not listening.

Fires when

Tyler is briefing, thinking out loud, or has just said stop or rejected a tool call — and a "let me just verify" call is forming. Also mid-way through an autonomous generation batch.

The rule

Stop instantly, with no verifying call first, and no plan-approval request mid-briefing. A rejected generation is a stop, not a pacing note: the next automated re-poke is not permission to resume. In autonomous generation, checkpoint after a batch rather than blasting batch after batch.

Cost when it fired

2026-07-14: "NO DO NOT VERIFY, AGGREGATE ALL PROBLEMS ALL MASTER AGENTS NEED TO BE ON THIS. STOP WHAT YOU ARE FUCKING DOING!!!!" — said while verify-subagents were being spawned mid-conversation, twice in a row.