The eval review page rendered by runes/core/skills/BuildSkill/eval-viewer/generate_review.py is functional but visually flat.
Restyle it using the report template the proton-ai-security project uses for its reports, keeping the current behavior unchanged: self-contained page, loopback-only server, per-case feedback auto-saved to feedback.json, --static output for headless environments, and model output rendered as untrusted text.
Template change only; no architecture change. The harness is slated to migrate into the cli (see the rune eval issue there), and the template moves with it.
The eval review page rendered by
runes/core/skills/BuildSkill/eval-viewer/generate_review.pyis functional but visually flat.Restyle it using the report template the proton-ai-security project uses for its reports, keeping the current behavior unchanged: self-contained page, loopback-only server, per-case feedback auto-saved to
feedback.json,--staticoutput for headless environments, and model output rendered as untrusted text.Template change only; no architecture change. The harness is slated to migrate into the cli (see the rune eval issue there), and the template moves with it.