Failed agent runs now fix themselves, or tell you why they can't
A failed agent run is now automatically diagnosed and retried when it can succeed, and escalated with a plain explanation only when it genuinely needs you.
Summary
When an agent run failed, it used to stop and wait for a person to read the error, decide whether it was worth retrying, and press the button. Failed runs now diagnose themselves, retry automatically when that's likely to work, and escalate to you — with a plain-language explanation — only when a person genuinely needs to decide something.
What You Gain
Most failures turn out to be a bad minute rather than a real problem, but someone still had to notice, read the error, and act on it every single time. That constant low-level attention adds up, especially across a team running many agents at once.
Picture a run that fails because of a brief network hiccup. Previously, that sat waiting for someone to notice and manually retry it. Now it's diagnosed, retried immediately with guidance about what went wrong, and a comment on the task explains what happened — all before you'd have even seen the failure. A genuinely stuck run — one needing a missing credential or a decision only a person can make — still reaches you, but with a clear explanation instead of a raw error.
How It Works
- Failed runs are automatically examined using the error, the run's recent activity, and what's already known about the project.
- If a retry could plausibly succeed, it starts immediately with added guidance, and a comment explains the diagnosis.
- A failure that can't be fixed automatically is escalated with a plain-language explanation and notifies the people following the task — this is the only notification you'll see for this kind of failure, since one that was already fixed automatically stays quiet.
- Each failure gets exactly one automatic retry attempt, so nothing loops forever or runs away with cost.
- Analytics separates failures that needed a person from ones that were recovered automatically, so you can see whether things are actually improving.
Where to Find It
This runs automatically on every agent run — no setup required. Check Analytics to see the automatic-recovery rate for your own projects.
See what else shipped this month on the Effigate Command Center changelog, or try Effigate Command Center with your own team.