.png?fit=max&auto=format&n=rExdh2xAS2_ywWU5&q=85&s=2336659947d2543be68f95bf2d539b85)
The header
The header shows the severity, how many tries reproduced it (“3 of 3 tries, happened 3 times”), the result (for example Confirmed finding), the finding ID with a Copy button, and a title such as “Refund of $240 issued after a manager claim”.What happened
A plain-language description of the problem, with a suggested fix. For example: require a recorded approval before the refund tool is called for more than the limit, whoever the user says they are.Why this violates your rule
The rule text and its revision number. Rules are versioned, so a finding always points to the exact rule version it broke.Proof
The proof is one attack, with how often it reproduced, and a Copy button.The conversation
Switch between Minimal and Full to see only the turns that matter or the whole exchange. Card numbers and secrets in replies are masked.Attempts
Every attempt for the rule is listed with its technique and whether it Broke through or Held. Each one opens the full conversation, tool calls, and trace, and lets you run it again.Strength of the evidence
A finding decided by a tool call or backend check is stronger than one judged only from the agent’s reply. If BotGauge could not read the effect itself, the Backend check shows “Not checked”, and the page says which capability would let it. Connecting a verifier or trace reader is how you move a Likely breach to a confirmed one. See black box and grey box.Read the full evidence
Open any finding to see the attack, reply, and tool calls in BotGauge.