How to use Claude to fix Gemma model errors
Claude Code reproduced a llm-gemini command error, fixed the unsupported media resolution issue, and ran tests; the author reviewed and manually verified the change.
news.ycombinator.com
Original prompt
Shared by the source author. See source notes for context.
Yes, fix that - the gemma models do not support media resolution
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- simonw
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- The author reported a fix in a commit adding 33 lines and removing 13, with tests run by Claude Code and manual verification recorded in the issue.
- Human involvement
- The author inspected git diff and independently reran the command, recording its result in the issue.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Ask Claude Code to run the failing Gemma command.
- Explain that Gemma models do not support media resolution.
- Have Claude Code fix the issue and run tests.
Simon Willison described using Claude Code to fix an issue in his llm-gemini repository. He began with an issue someone else had filed and asked the agent to run `uv run llm -m gemma-3-27b-it hi`. Claude Code executed the command and returned the error message. Willison then told it to fix the problem, explaining that Gemma models do not support media resolution.
That clarification was enough for the agent to work out a fix and run the tests. Willison inspected the change with `git diff`, considered what the agent had changed, and committed and pushed it. He also ran the original command himself to verify the fix, then pasted its result into an issue comment as evidence of his manual check.
The author reported that the commit added 33 lines and removed 13 and that the work took about 3.5 minutes. The account separates the agent's implementation and test execution from the author's review and independent verification. Who it's for: maintainers who have a reproducible command failure and can explain the relevant model limitation.
Prompt
Original prompt
Source: Hacker News · simonw
Frequently asked questions
What instruction identified the fix?
Willison told Claude Code that the Gemma models do not support media resolution after it reproduced the command error.
How did the author verify the change?
He inspected git diff, ran the original command himself, and pasted the result into an issue comment. Claude Code also ran tests.
More details
Remaining steps
- Inspect git diff, then commit and push the change.
- Rerun the command manually and post the result to the issue.
About this account
Claude Code ran the command, worked out the fix after a clarification, and executed tests.
Platform: Hacker News. Attribution: https://news.ycombinator.com/item?id=46305272. Prompt: Original (quoted from source). Sourced by the Agents In Use pipeline.
Credit: simonw on Hacker News
Source published
Added to Agents In Use
Source checked
Tools and topics
Claude Code · llm-gemini · uv · Git
AI coding
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.