<< All versions

Skill v1.0.0

currentAutomated scan100/100
tiger-ai-lab/rewardharness/evaluating-multi-step-edits
──Details
PublishedNovember 3, 2026 at 02:38 AM
Content Hashsha256:ae31a68ffeffe1bb...
Git SHA0c1e7051ab4f
──Files
Files (1 file, 974 B)
SKILL.md974 Bactive
SKILL.md · 18 lines · 974 B

schema_version: 2 name: evaluating-multi-step-edits kind: skill description: Guidelines for evaluating prompts with multiple distinct instructions. metadata: version: "1.0.0"


Evaluating Multi-Step Edits

When a prompt contains multiple instructions (e.g., "Remove the object on the left AND increase the brightness of the central figure"):

  1. Deconstruct the Prompt: Break the prompt down into its individual components.
  2. Check Each Component: Evaluate whether each edited image successfully completes every component.
  3. Partial Success: If an image only completes one of the two instructions, it should receive a lower Instruction Following score (e.g., 2 or 3) compared to an image that completes both.
  4. Over-editing: Ensure that global adjustments (like brightness) are only applied to the specified areas (e.g., "central figure") and not the entire image, unless requested.

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

All versions