Codex fixes Markdown leaking into social previews
Simon Willison gave Codex a specific defect in his Django blog: social preview descriptions exposed raw Markdown. Codex changed templates and added a passing test in a configured PostgreSQL environment.
simonwillison.net
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
In this repository, fix [specific visible defect]. First reproduce it with the existing test runner. Make the smallest change, add a regression check, and report the diff and test output. Work on a branch; stop before merging or publishing.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source author
- Simon Willison
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- Willison reports a template fix and a passing regression test; the published preview no longer shows Markdown.
- Human involvement
- Willison configured the environment, disabled agent internet access and inspected the resulting change.
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
- Configure the repository container with PostgreSQL and the blog dependencies.
- Ask Codex to strip formatting from preview descriptions and include a test.
- Let it edit the templates and run checks, then inspect the resulting social preview.
More details
About this account
Codex selected the code changes and executed the test loop after receiving the bug and environment details.
Firsthand account links the patch and agent session. Reported result, not an independently reproduced benchmark.
Credit: Simon Willison
Source published
Added to Agents In Use
Source checked
Tools and topics
Codex · GitHub · Django · PostgreSQL
Code · Testing
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.