Codex silently truncates AGENTS.md at 32 KiB, developers find
10/21/2026 — 10/21, 12:19·1 sources (all)·1 reports (all)
Key takeaways
- A developer's testing shows that Codex only parses the first 32 KiB of AGENTS.md, silently truncating the rest by bytes with no warning;
- subdirectory rules are lost first, and even OpenAI's own repo test guidelines went unread. The write-up compares the behavior with Claude Co…
Background & analysis
On October 21, 2026, a report published on Juejin described a developer's hands-on testing of how Codex handles AGENTS.md. According to the report, Codex parses only the first 32 KiB of the file and silently truncates anything beyond that byte limit, without showing any warning. Rules placed in subdirectories are said to be the first to disappear once the cutoff is reached, and the developer also found that the test specifications in OpenAI's own repository were not picked up.
The report included a side-by-side test against Claude Code, along with a self-check command that developers can use to determine whether their own rule files are being truncated. As of the report's publication, the matter remained at the stage of developer testing and comparison; the material contains no response from OpenAI and no indication that a fix has been made.
AI-generated from 1 reports · updated 13 hours ago
Latest turnA developer's tests show Codex parses only the first 32 KiB of AGENTS.md, silently truncating the rest by bytes with no warning, so rules in subdirectories are dropped first — even OpenAI's own repo test guidelines go unread. The post includes a side-by-side test against Claude Code and a self-check command to verify whether your own rules file is being cut off.
Related AI tools
- Heat index
- 67
- All sources
- 1
- All reports
- 1
- First seen
- 15 hours ago
Reports on this story headlines open the original
A developer's tests show Codex parses only the first 32 KiB of AGENTS.md, silently truncating the rest by bytes with no warning, so rules in subdirectories are dropped first — even OpenAI's own repo test guidelines go unread. The post includes a side-by-side test against Claude Code and a self-check command to verify whether your own rules file is being cut off.
掘金AI score 62
Other stories people are talking about
- Heat index 76Developer open-sources voice-controlled projector app powered by Codex1 sources
- Heat index 96Developer pins Claude Code and Codex model tiers by task type1 sources
- Heat index 85Developer Open-Sources ccx Tool for Synchronizing Context Between Claude Code and Codex1 sources
- Heat index 450NewOpenAI Announces Integration with Google Cloud, Bringing Models and Codex to Model Garden5 sources
- Heat index 86Developer Practical Experience Running Codex in Real Spring Boot Codebases1 sources
How is heat calculated?About the methodHide
Heat counts how many independent sources covered a story in the last 48 hours: one source counts once no matter how many posts it published, decaying with a 24-hour half-life. What ranks first is what many people are talking about.
This page aggregates public feeds. Headlines and summaries are machine-organized and remain the property of the original authors; verify important facts at the source. If you believe a headline or summary infringes your rights, email the contact address in the footer with the page URL and basis of your claim, and we will remove or replace it after review.
- Surge
- Discussion rising fast
- New
- First report within 6 hours
- Rising
- Still gathering discussion