AIToday
Large Language ModelsAI Coding AssistantsZenn AI/MLPublished: Oct 9, 2026, 10:00 JST

Opus 5.5's test edge was repos, not the model: developer

Opus 5.5's test edge was repos, not the model: developer

A developer logged 80 commits across five WordPress plugin repos from September 1 to October 3, all co-authored with Claude, spanning Claude Opus 4.8, Claude Opus 5 and Claude Opus 5.5.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, in your inbox every morning.

Context & Analysis

The workflow behind these commits stayed the same as the models swapped. Bug reports arrive through the WordPress.org forum or email, get brought into a claude.ai chat to have the code read, and are written up as a handover document covering cause and fix. That document goes to Claude Code, which implements the change and produces the commit. In one September 15 example, a forum report that a reference card under a chat answer pointed to an unrelated page was traced to vector search results not being flagged, producing a 1.19.5 commit co-authored with Opus 4.8 the same afternoon.

The handover document itself followed a fixed shape: symptom, the file and line where the cause sits, what to fix and what not to fix, verification items, and a draft changelog line. The developer suggests this uniform input may be why the commit format looked so consistent — 60 of 80 commits updated readme.txt in the same commit, and 50 of 80 subjects began with a version number — while noting this is a guess that was never compared against not using a handover document.

What the record cannot show is where human judgement ended and the AI's began. The co-author line only signals AI involvement. On September 30, PDF Image Creator shipped eight releases in one day around image regeneration work, and how the team decided to split that into eight is not captured. Pricing and deadline calls, plus real-server checks such as hitting the Stripe API through PowerShell or inspecting ImageMagick on a rental server, happened outside the commits and left no trace in the repositories.

FAQ
Which Claude models appear in the commits?
Three: Claude Opus 4.8 (9 commits, 9/1–9/15), Claude Opus 5 (47 commits, 9/3–9/22) and Claude Opus 5.5 (24 commits, 9/24–10/3). The model name in the co-author line changed three times over the period.
Why did Opus 4.8 update no tests?
All 9 of its commits landed in the two repositories that have no tests/ directory at all, so there were no tests to update.
Was the model the only variable compared?
No. The developer notes the 80 commits cannot separate model effects from repository effects, because the same repository was never compared at the same time with only the model changed.

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleHex cuts wrong edits 7x with evals on Quick Edits