Verified:GitHub

v14-eval-ops

Designs evaluations, scorecards, and release gates for prompt and agent quality control.

Updated May 22, 2026

Search Tags:#drift detection #evaluation #regression testing #llm-as-a-judge #prompt quality #scorecard #release gating

One-click install

npx skills add https://github.com/handaeho/Harness_Engineering --skill v14-eval-ops-handaeho

Or copy as Structured Prompt for Agent▼

Please help me install this Agent Skill.
Skill: v14-eval-ops
Source: https://github.com/handaeho/Harness_Engineering/tree/main/prompt-stack/_legacy/_cold_storage/legacy_older_versions/v14/codex/skills/v14-eval-ops
Command: npx skills add https://github.com/handaeho/Harness_Engineering --skill v14-eval-ops-handaeho

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Teams shipping prompts, agents, or workflows often lack a disciplined way to compare versions, detect regressions, and decide whether a candidate is release-ready, leading to decisions based on vibes rather than evidence. ## Core Features & Use Cases - Evaluation Design: Select the right evaluation surface, define contract-first acceptance criteria, and build metric taxonomies tied to actual decisions. - Regression and Drift Analysis: Compare baseline versus candidate behavior, detect semantic drift, coverage regression, and anomalies without being misled by noisy single points. - Release Gating: Define gates with owners, thresholds, and actions for rollout, canary continuation, and prompt-version promotion. - Use Case: When comparing two prompt versions before rollout, use this Skill to build a scorecard with weighted criteria, identify critical-cohort failures, and produce a release recommendation with explicit evidence and unresolved risks. ## Quick Start Use the v14-eval-ops skill to compare the baseline and candidate prompt versions and produce a release-readiness scorecard with regression findings.

Frequently Asked Questions about v14-eval-ops

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

Why do averages hide serious evaluation failures?▼

Averages can mask severe failures in small but critical cohorts, such as safety regressions affecting a minority of cases. Surface critical-cohort failures separately and treat safety regressions more strictly than cosmetic ones.