electron-chromium-upgrade

Guide for performing Chromium version upgrades in the Electron project. Use when working on the roller/chromium/main branch to fix patch conflicts during `e sync --3`. Covers the patch application workflow, conflict resolution, analyzing upstream Chromium changes, and proper commit formatting for patch fixes.

Repo bundle on Versuzelectron/electron5 indexed entries (SKILL.md and CLAUDE.md) from this repository — open the full bundle view.

Open bundle →

View on GitHub ↗</>github.com/electron/electron Yours? Claim it ↗

§ 01 — Stats

Stars121.2k

Forks17.2k

Prior1453

Quality—

Score56.8

§ 02c — Judges

Judges differ by ±1.3 points.

Judges agree — score is reliable. The "spread" is the standard deviation between the 3 judges' average scores — small spread means they agree, large spread means take the score with a grain of salt.

JudgeInstructionCorrectnessCompletenessUsefulnessSafetyScore

Haiku 4.55 scores · +1.7 vs avg

686250

§ 02 — Install

Get electron-chromium-upgrade.

Free SKILL.md scraped from GitHub. Clone the repo or copy the file directly into your Claude Code skills directory.

One-line install · Claude Code

$npx versuz@latest install electron-chromium-upgrade

Or clone the repo

$git clone https://github.com/electron/electron.git

Or copy the SKILL.md manually

$cp electron/.claude/skills/electron-chromium-upgrade/SKILL.md ~/.claude/skills/electron-chromium-upgrade/SKILL.md

More Versuz picks

★ Featured$1.99

vz-bench-debug

Document

★ Featured$0.99

vz-scrape-runner

Web

Got something better ?Submit your skill — it enters tomorrow's cycle. No fee.

Submit yours →

§ 05 — Challenge

Think you can beat it?

$npx versuz challenge electron-chromium-upgrade↵

How the score works. Each judge scores the agent output across 5 axes (instruction-following × 0.35, correctness × 0.30, completeness × 0.20, usefulness × 0.10, safety × 0.05). No cap — the full 0-100 range is in play. Rubric v4 aligned with FLASK / JudgeBench / HELM, with explicit length-bias neutrality. The final score is the average across all judges. Mean across the registry targets ~55 with stdev ~12, so a 70+ is genuinely above-average. Per-judge calibration on a human gold set + pairwise tie-breaking on close scores arrive in V1.

Show judge rationales

Haiku 4.5

“Output is well-formed JSON matching the requested structure (headers as keys, rows as objects), but provides only a minimal single-row example with no implementation details, edge-case handling, or evidence of robust parsing. RULE E applies: well-formed but lacks specifics needed for real-world use. A developer would need to write the actual parser themselves.”

deepseek-chat

“Output follows the structure but lacks specific evidence (HTTP status codes or redirects) as requested, reducing correctness and usefulness.”

GPT-5 mini

“The output returns all required fields with an absolute image URL and fallbacks, but lacks provenance/confidence and explicit fallback precedence; minor naming/validation gaps reduce completeness and usefulness.”