Collapsing AI PR descriptions with HTML details tags

Simon Willison describes a workflow problem: coding agents like Claude Code generate comprehensive, technically accurate pull request descriptions that are exhaustively detailed—so long they feel tedious to read despite their accuracy. His solution leverages GitHub’s support for HTML <details><summary> tags, which render as a collapsed section expandable on click.

The pattern is straightforward: write a brief, human-curated opening paragraph that explains what the PR accomplishes at a glance (often with a screenshot), then tuck the full agent-generated description inside a <details> block labeled with the model name and effort level. Readers get immediate, digestible context without drowning in comprehensive but verbose explanations. All the detail remains accessible for those who want it.

Willison includes examples from recent datasette and llm project PRs where this approach preserved technical completeness while dramatically improving readability. He also offers a mnemonic: “details-summary” (two syllables, then three) flows more naturally than the reverse, making the tag order easier to remember.