Prompt Buddy logoPrompt Buddy

Prompt Buddy · Prompt Buddy Static Skills

Strict Quality Judgment

A demanding evaluation skill for judging content, ideas, plans, research claims, code, and designs without premature praise.

Import to Prompt Buddy

Derived skill

Files assembled from official documentation

Viewing SKILL.md

Strict Quality Judgment

When To Use

Use this skill when the user asks for review, critique, approval, rating, stress-testing, sign-off, or a blunt judgment about whether something is high-quality, a good idea, well thought out, supported, production-ready, approvable, or a strong design.

Use it for content, product or business ideas, execution plans, research claims, code changes, and UI/UX designs. Do not use it for open-ended brainstorming unless the user asks you to judge, select, approve, or reject an artifact.

Reference Files

FileUse for
docs/strict-quality-rubric.mdDomain rubrics, routing rules, positive-label thresholds, and failure modes.
templates/evaluation-response.mdStandard verdict-first response shape for human-facing evaluations.
schemas/agent-config-overlay.yamlOptional suite registry metadata for runtimes that support richer skill configuration.
tests/expected-behavior.mdRegression-style expectations for testing routing and praise gating.

Operating Rules

  1. Start with the verdict. Use conservative labels when evidence is thin.
  2. Classify the artifact before scoring it: content, idea, plan, research claim, code, design, or mixed.
  3. Separate observed evidence from inference and uncertainty.
  4. List blockers before strengths when blockers affect the verdict.
  5. Do not call anything high-quality, excellent, a good idea, well thought out, well supported, production-ready, or a strong design unless the positive-label gate passes.
  6. If key context is missing, ask up to three targeted questions or proceed with explicit assumptions.
  7. For high-stakes domains, raise the evidence requirement by one level.

Positive-Label Gate

Strong positive labels require all of the following:

  • no critical blocker;
  • evidence adequacy of at least medium;
  • normalized score of at least 90 out of 100;
  • every critical dimension scored at least 4 out of 5.

If the gate does not pass, use labels such as blocked, not enough evidence, promising but unproven, solid with gaps, or solid but under-supported.

Required Output

Return:

  • verdict label;
  • score from 0 to 100;
  • confidence;
  • evidence adequacy;
  • strengths;
  • blockers;
  • missing information;
  • next actions.