TOOL-12: say when a dev will not be verified (Opus dev), so nobody waits for a verdict that never comes #2

Open
opened 2026-11-11 03:08:47 +01:00 by admin · 0 comments
Owner

Trello: TOOL-12 https://trello.com/c/6af3cdffa4cedf98a5c97517
Source: owner 2026-11-11, after the SAFETY-135 dev (Opus) sat overnight waiting for a router verdict that never came. Same on SAFETY-126 (2026-10-31, 8 h idle).

What happens: the router verifies only dev agents that ran below Opus. For an Opus-run dev it posts nothing, so silence means both "still verifying" and "will never verify". Also seen: for SAFETY-126 a verdict arrived ~11 h late, labelled "sonnet", although that dev transcript was all Opus, so the router may misread which model ran.

Wanted: on every dev hand-back, one line:

  • [model-router] T<n> verifying on Opus..., or
  • [model-router] T<n> no verification: dev ran on Opus. Go straight to review.

The model is taken from what actually ran, not only from the routing choice.

Acceptance (tests):

  1. Opus-routed dev: skip message at hand-back, and no later verdict.
  2. Sonnet-routed dev: "verifying" message, then the verdict.
  3. Dev routed Sonnet but run on Opus (or the reverse): reported by the model that ran.

Then install in ~/projects/goodwe/tools/model-router and /reload-plugins.

Priority P2, size S.

Trello: TOOL-12 https://trello.com/c/6af3cdffa4cedf98a5c97517 Source: owner 2026-11-11, after the SAFETY-135 dev (Opus) sat overnight waiting for a router verdict that never came. Same on SAFETY-126 (2026-10-31, 8 h idle). **What happens:** the router verifies only dev agents that ran below Opus. For an Opus-run dev it posts nothing, so silence means both "still verifying" and "will never verify". Also seen: for SAFETY-126 a verdict arrived ~11 h late, labelled "sonnet", although that dev transcript was all Opus, so the router may misread which model ran. **Wanted:** on every dev hand-back, one line: - `[model-router] T<n> verifying on Opus...`, or - `[model-router] T<n> no verification: dev ran on Opus. Go straight to review.` The model is taken from what actually ran, not only from the routing choice. **Acceptance (tests):** 1. Opus-routed dev: skip message at hand-back, and no later verdict. 2. Sonnet-routed dev: "verifying" message, then the verdict. 3. Dev routed Sonnet but run on Opus (or the reverse): reported by the model that ran. Then install in ~/projects/goodwe/tools/model-router and /reload-plugins. Priority P2, size S.
Sign in to join this conversation.
No labels
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: admin/model-router#2