I have text begin returned from a query that is a tool in an Agent. However, some of the text is being redacted as there is presumably some sort of match with a secret configuration variable.
Can this behaviour be turned off, as these are false positives?
Hey @klautier - I've been seeing similar reports about redacted text in various parts of the product and will do some further investigation. I'll provide an update here as soon as I can!
When you say that this is a false positive, do you mean the redacted text doesn't actually match an existing secret or that the match is just a coincidence? I'm trying to understand whether there's something buggy happening or if we need to implement a toggle. It doesn't look like we've made any recent changes to this part of the codebase, so I'm guessing it's the latter.
In general, though, your understanding of the intended functionality is spot on. Anything being returned to the client after being evaluated by the workflow backend - which includes our Agent orchestrator - is automatically scrubbed of all secrets.
Sorry, to be clear, the matches are coincidental, i.e. the text matches and so is redacted, but the redacted text was not itself anything special and not secret.
