wassnameandClaudypoo 7d5342f332 prompts: varglite evidence discipline -- quote what you observed, judge rejects reconstructed quotes ...

Dogfood: an agent with blank tool output back-filled plausible test counts into evidence and
the judge accepted (the numbers happened to be true). Norm now stated agent-side (verbatim
quotes, honest gaps beat plausible fabrication) and enforced judge-side (mismatched quotes =>
reject even when the goal looks met).

Co-Authored-By: Claudypoo <288921227+claudypoo@users.noreply.github.com>

2026-07-03 12:04:04 +08:00

..

index.ts

nudge when plan.md exists but no goal line matches, instead of going silently inert

2026-07-03 12:00:01 +08:00

prompts.ts

prompts: varglite evidence discipline -- quote what you observed, judge rejects reconstructed quotes

2026-07-03 12:04:04 +08:00