<< All versions

Skill v1.0.0

currentAutomated scan100/100
tiger-ai-lab/rewardharness/evaluating-thematic-and-setting-accuracy
──Details
PublishedNovember 2, 2026 at 05:57 PM
Content Hashsha256:3874209db0ab711a...
Git SHA0c1e7051ab4f
──Files
Files (1 file, 769 B)
SKILL.md769 Bactive
SKILL.md · 14 lines · 769 B

schema_version: 2 name: evaluating-thematic-and-setting-accuracy kind: skill description: Guidelines for evaluating full setting transformations and ensuring exact thematic matches. metadata: version: "1.0.0"


When evaluating full setting transformations (e.g., 'elementary school cafeteria', 'sold out concert', 'underground New York subway'), look for specific distinguishing elements. A classroom with desks is NOT a cafeteria. A room with a few people is NOT a sold-out concert. A generic train station is NOT a New York subway. If the edited image retains too much of the original setting (e.g., keeping a restaurant table when asked for a school cafeteria), penalize its instruction score.

<!-- Release review: RewardHarness v0.4.1 (2026-08-26). -->

All versions