Use swarms to find bugs through property tests
A developer ran Codex swarms against code properties and turned up about 30 bugs in company code and around 100 more in open-source projects.
- What they did
- A developer gave a talk at work about property testing with agent swarms, then reused that idea on outside projects so their article would not center their employer’s code. They sent the agents at projects such as uv, codex, and prometheus. The announcement leaves out the operational details of how the runs were carried out.
- How it went
- They say the effort exposed roughly 30 bugs in their employer’s code and about 100 more across the open-source targets. Many fixes are already merged, but the post does not provide a project-by-project tally.
- Tips
- This approach is most useful when software behavior can be checked against general rules across many inputs. Tell the agent not to edit, send, or submit anything until you review the failures and give permission.
Try it yourself with Codex
Take [repo/module] and look for behaviors that should hold across many inputs, especially in parsing, serialization, validation, and data transformation code. Generate property-based tests using [your testing framework], run them, and collect any failures. For each failure, shrink it to the smallest reproducible case and explain whether it looks like a genuine bug. Do not alter the source, file an issue, send a patch, or submit anything until you stop and ask me first; only proceed after I approve. Finish with a short report of the verified bugs, how to reproduce them, and the most likely fix for each one.
Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.
Get 3 agent uses like this every morning
With the day's top AI news. Free, and one click to unsubscribe.
Related agent uses
Investigate why a service is getting killed
Run a read-only SSH check to find the cause of a process death and rank fixes by risk.
Move Claude Code skills into Codex
Charlie Hills used Claude Code and Codex to migrate 201 skills in an afternoon, then checked which parts worked in both tools and which still needed fixes.
Turn tracked tickets into pull requests
Use an agent to pick up issues or tickets, make the code change in an isolated pod, and open a PR for review.
Use a long-running agent to chase a stubborn bug
A developer let Codex work overnight against a repeatable test, using debug prints and hypotheses to track down an unexplained bug.
Use an agent to patch a stubborn app bug
They pointed a coding agent at two annoying software behaviors and had it debug one and add a self-improvement feature to the other.
Create and launch a BaZi reading agent
A developer used coding agents to build a BaZi reader that calculates charts, writes interpretations, and handles language detection and USDT payments.