An agent made the same call again and again
What you're seeing
A finding like:
Refund bot made the same call 9 times in a row: issue_refund on pay_3Nq in razorpay, over 40 seconds, with nothing different in between. Outcomes: 0 succeeded, 9 failed, 0 not recorded.
It is a Same call repeated finding, one of the early warnings. It is medium when every call in the run failed, and low when some did not.
What it means
In one trace (or one session, when there is no trace), consecutive records of one agent had the same action type, tool and target, each within five minutes of the one before, five or more in a row. Anything different in between ends the count, so an agent that alternates two calls is not reported by this finding.
What it cannot tell you is why. A stuck loop, a retry policy that keeps trying a failing call, a poll, and a pagination that does not change the target all look the same in the records. The finding says "repeated" and not "stuck" because only the owner knows which it is.
What is not counted:
- Model calls. An agent reasons through many model calls in an ordinary run.
- Calls with neither a tool name nor a target. They are too anonymous to say they were the same call.
- Calls in a trace or session with no identifier, because there is no way to say which calls belong together.
What to do
- Open the first and last call from the finding. Look at the outcomes and at what the target is. Nine failures against one target is usually a retry that should have given up.
- If the agent is stuck, stop or fix it in the system that runs it. Provantix cannot stop it: it reads records afterwards. The cost and volume findings on Home show whether it is also spending.
- If the repetition is expected, say how often the agent runs: open the agent, choose Expected to run and set the interval. An agent with a declared interval is not reported for repeated calls that are a minute or more apart, because that is its schedule. Calls faster than that are still reported.
- Resolve the finding with the cause if you want it on the record. It closes by itself when the run is more than a day old.
The numbers
| Calls in a row | 5 or more |
| Longest gap between two calls in a run | 5 minutes |
| Looked at | The last 24 hours |
| Scheduled agent | Skipped when the middle gap in the run is a minute or more |
What this does not do
It does not stop anything, does not judge the agent, and does not look at what the calls returned. It reports that the same call happened, how many times, over how long, and with what outcomes.