update-golden-values

Refresh golden test values from GitHub Actions workflow runs and generate PR-ready relative-difference summaries.

17.4k|4.3k|Updated Mar 21, 2019
One-click install
npx skills add https://github.com/NVIDIA/Megatron-LM --skill update-golden-values-nvidia
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: update-golden-values
Source: https://github.com/NVIDIA/Megatron-LM/tree/main/skills/update-golden-values
Command: npx skills add https://github.com/NVIDIA/Megatron-LM --skill update-golden-values-nvidia

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

It refreshes failing or stale golden test artifacts from a specified GitHub Actions workflow run and quantifies how much the updated goldens differ from what is currently checked into the repository.

Core Features & Use Cases

  • Golden value refresh from workflow artifacts: Downloads and overwrites golden value JSON files produced by a given workflow run, with optional scope limited to failing/cancelled jobs.
  • Per-metric relative-difference scoring: Computes a per-(file, metric) average normalized relative difference (avg_rel_diff) over shared steps and ranks changes by magnitude using absolute relative diff.
  • PR-ready narrative output: Produces a structured summary that can be pasted into a pull request description, highlighting the metrics and buckets with the largest movement.

Quick Start

Run the update for a workflow run ID by refreshing only failing goldens, then paste the generated summary into your PR description: set the GitHub token via your authenticated gh CLI and execute the download and comparison flow using the workflow run ID you provide.

Frequently Asked Questions about update-golden-values

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

FAQPage Schema
How do I update golden test values from a GitHub Actions workflow run?

Golden test regression triage analyzes failing CI workflow artifacts to identify which specific metrics diverged from the baseline. It computes a per-file, per-metric average normalized relative difference over shared steps to quantify the movement and rank changes by magnitude.

How do I generate a PR-ready summary for golden value changes?

Yes, you can limit golden value refresh to failing jobs only. The tool applies to CI workflows where users need failing-only or full golden updates, allowing you to scope the artifact download and comparison process specifically to cancelled or failing workflow runs.

What do I need to download CI artifacts and refresh goldens?

Relative diff scoring in golden tests computes a per-(file, metric) average normalized relative difference over shared workflow steps. It ranks changes by absolute relative diff to precisely quantify how much updated goldens differ from the checked-in repository baseline.

When should I refresh failing goldens instead of updating all test values?

Golden value regression triage identifies which specific metrics diverged from the repository baseline during a CI workflow run. It computes a per-file, per-metric average normalized relative difference over shared steps to quantify movement and rank changes by magnitude for review.