# Simon Willison hides his agents' PR descriptions in a collapsed block, and writes the first line himself

> Agent-written pull request descriptions are accurate but read like slop. His fix: a short human opener, with the full write-up tucked into an HTML details tag labelled with the model.

- URL: https://devisdead.ai/news/simon-willison-hides-his-agents-pr-descriptions-in-a-collapsed-block-and-writes-
- Published: 2026-11-09T12:42:01Z
- Author: DevIsDead Newsroom
- Section: Dev
- Tags: pull requests, coding agents, github, workflow
- Originally reported by Simon Willison's Weblog (https://simonwillison.net/2026/Nov/9/details-summary/).

Coding agents write excellent pull request descriptions. That's the problem. Simon Willison explains on [his weblog](https://simonwillison.net/2026/Nov/9/details-summary/) why that has become a readability issue, and the small trick he now uses.

## Too good to read

Most of Willison's changes now arrive as PRs opened for him by Claude Code or Codex. The agent-written descriptions are accurate and thorough: the problem, the solution, how it was verified. But that's a lot to take in, and it reads like AI writing even when it isn't wrong. His gut reaction, he admits, is "I'm not reading all of that", even when he was the one driving the agent.

What he wants is human editorial judgment: someone who thought about the PR just long enough to say, in a sentence or two, what it's for and whether it deserves attention.

## The fix

GitHub Markdown supports the HTML `details` and `summary` tags, which render as a collapsed heading you can expand. Willison now opens each PR with a short, carefully chosen explanation, sometimes with a screenshot, then puts the agent's full write-up inside a collapsed block. The summary label is the model and often the effort level, like "Claude Opus 5.5 Medium", so it's obvious what's inside.

Sometimes he writes the opener himself; sometimes he lifts the useful sentence out of the agent's text. Either way, his promise to readers is that the short version reflects his own judgment about what's worth their time.

His examples include a Datasette PR whose opener simply notes that the row page used to scroll far to the right for wide tables, followed by a screenshot.

## Why it's worth copying

It's a tiny, practical answer to a growing problem: agents generate more text than humans want to read, and reviewers start skimming everything. Separating the human summary from the machine detail keeps both, and keeps the reviewer's trust. It also makes clear which parts a person actually stands behind.

**[Scoreboard](/scoreboard) update:** Humans +1. The agent writes the essay; the human keeps the job of saying what matters.

Stay undead.