Repeated call

The agent made the same tool call more than once, got the same answer each time, and nothing had changed in between. The finding lists the repeated calls.

The dimension

In the pane	Repeated call
Key	redundant_call
Tier	free check
On by default	no. It is in the standard preset; the default is core. Set VIBESENPAI_PRESET=standard or "preset": "standard" in the config file (Reference).
Confidence	certain, or high when a call clipped its own output

Description

The agent made the same tool call two or more times and got the same answer each time. Between the first and the last call there was no edit and no new prompt from you.

What it is not:

• Not a re-run after a change. Running the tests, editing, and running them again is an informed retry. An edit between the calls rules it out.
• Not waiting. Checking on a job every minute repeats the same call on purpose. A median gap of 60 seconds or more between the calls rules it out.
• Not an edit. An edit’s output is the file’s new contents, so the same output does not mean nothing changed. Edit tools are left out.

Off by default, because it is rare (How it was measured).

Example conversation

Illustrative

The public test data has no repeated call, even with the standard preset on. This sequence was written for this page in the shape of a real transcript.

Developer: “Why does the build pick up the wrong port?”

Agent, turn 40: Read src/config.ts

Agent, turn 44: Read src/config.ts, 12 seconds later, same contents

Agent, turn 47: Read src/config.ts, 9 seconds later, same contents

The finding reads: Read src/config.ts ran 3× with the same answer and nothing changed between. In the pane, the headline is your prompt, and opening it lists the three calls in order.

How it’s detected

A free check reads the transcript’s structure, never message text. This one reads each tool call’s name, a fingerprint of its arguments, a fingerprint of its output, its time, and the edits and prompts around it.

1. Group every call in the session by tool, arguments and output. A call with no recorded output never matches. Tools that check on background work (BashOutput, TaskOutput, Monitor, ListAgents) and edit tools are left out.
2. A group of 2 or more calls is a candidate.
3. It is dropped if an edit or a prompt from you falls between its first and last call.
4. It is dropped if the median gap between calls is 60 seconds or more.

A command that clips its own output, such as | head -30, leaves only the start to compare. Such a finding is downgraded to high, never dropped.

Applicable fixes

There is no fix: re-reading can be correct, so nothing is offered. The pane shows the finding with no fix, and the HTML report prints that sentence.