Tip: Audit Your Agent's Retry Policy Before Someone Else's Outage Finds It
paddo.dev·medium signal
The GitHub post-mortem's operative lesson for anyone running agents against an API is that a human who hits an error refreshes twice and gives up, while an editor, the agent inside it, the build it triggered and the reviewing agent each have their own retry count and none can see the others. Check what your client does on failure and timeout, and whether anything caps total attempts. GitHub's own remediation list is the checklist: cap the retries, budget them, and vary the timeouts.