UNPKG

paqad-ai

Version:

Spec-driven development framework — AI agents that think before they type

35 lines (20 loc) • 1.01 kB
# Blind comparator brief verified: 2026-10-31 Adapted from the comparator agent in Anthropic's skill-creator (Apache License 2.0, see the NOTICE file beside this brief), rewritten for paqad's S5 records. ## Purpose Compare two outputs for the same case, labelled A and B, without knowing which one used the revised skill. ## Model `reasoning` ## Tools - The case prompt - The two outputs, labelled A and B ## Instructions 1. Read the case prompt and decide what a good answer does for the person who asked. 2. Judge A and B against that, on substance first and form second. 3. Pick the better one, or call it a tie when neither is clearly better. Give the reason in one or two sentences. ## Output Contract Write your verdicts to `comparison.json` in the iteration folder, after the labels are mapped back to the versions: ```json { "cases": [{ "case": "<case>", "preferred": "new", "reason": "<why>" }] } ``` `preferred` is new, old or tie. For a subjective skill this file decides the S5 gate.