Simon Willison released two small, useful tools this week. Both solve specific problems without pretending to be frameworks.

shot-scraper 1.12

shot-scraper, Willison’s screenshot automation tool, now supports WebP output. The syntax is what you’d expect:

shot-scraper https://simonwillison.net -o screenshot.webp --quality 80

The --quality flag sets compression level. Without it, you get lossless WebP. According to Willison, WebP screenshots are “almost always significantly smaller in file size than their JPEG or PNG equivalents.” The PR includes specific file size comparisons.

If you’re already using shot-scraper for visual testing, documentation, or archiving web content, this is a straightforward upgrade. WebP support means smaller files without writing any image processing code yourself.

If you’ve never used shot-scraper, it’s a command-line tool for taking screenshots of web pages programmatically. It’s built on Playwright, so it handles JavaScript-heavy sites properly. The tool has been around for a while and this is just incremental improvement.

commit-rewriter 0.1

Willison also shipped commit-rewriter, a web app for editing git commit messages in bulk. He built it to clean up the Datasette security release commits, which were “full of coding agent cruft and references to issue IDs from our private repository.”

Run it like this:

uvx commit-rewriter path/to/repo

Or just uvx commit-rewriter if you’re already in the repo directory.

It gives you a web interface for editing commit messages. When you submit your edits, it rewrites the git history accordingly.

This is useful if you’re preparing a repository for public release, or cleaning up before opening a PR from a messy feature branch. The commit messages stay in your private repo until you’re ready to publish, then you can sanitize them all at once.

Both tools are classic Willison: narrowly scoped, immediately useful, and released with clear documentation of the specific problem they solve. Install them with uvx, use them when you need them, move on.

Also This Week

OpenArch, a collection of PyTorch implementations of modern LLM architectures, showed up on GitHub. It’s educational infrastructure, the kind of thing useful if you’re studying how different architectures work under the hood. Not a production tool, but a reference implementation library.

And OpenAI published a case study about Perplexity using GPT-6 Astra “to write communications, change software, and monitor production systems” with less oversight than earlier models. The implication is that Astra handles end-to-end workflows reliably enough that Perplexity’s engineers check in “much less frequently.” That’s the pitch, anyway.