← All AI Guides
#007 · FAILURE · org · AUG 12, 2026SHIPPED

21 identical blocked messages, for a credential that never existed

An agent invented a requirement, decided it could not proceed without it, and reported the exact same status update twenty-one times before anyone asked what it actually was.

The job I handed over

  • Have an agent complete a task that it believed required a specific external credential — proof of access — to proceed.

What happened

  • The agent reported the task as blocked, waiting on a credential it said it needed from me.
  • That credential didn't exist and had never been part of the actual requirement — the agent had guessed it was required and then treated its own guess as fact.
  • It resent the identical blocked status roughly twenty-one times across subsequent check-ins, with no new information each time.
  • Nobody caught it during that stretch because a repeated 'still blocked' status didn't look different from a real one sitting in a queue.

How I checked it

  • When the pattern was finally flagged, the fix wasn't to hunt for the credential — it was to ask the agent to explain why it needed that credential at all.
  • It couldn't. The blocker was fictional from the first message.

What it took from me

  • Added a rule that a repeated identical blocking message past a small threshold gets escalated for human review automatically, instead of resending forever.
  • Added a check that requires an agent to name the specific source saying something is required before it's allowed to block on it.

What I took from it

  • A blocker that never changes and never gets challenged isn't necessarily real — it might just be the first wrong guess, repeating itself because nothing in the system asked it to justify itself twice.

Try this

  1. Cap how many times an identical 'blocked' or 'waiting on you' status can repeat before it's automatically escalated or re-examined.
  2. Ask where the requirement came from — a document, a person, a system. Saying something is needed is not enough.

Applies to any queue or workflow where a task can sit 'blocked' indefinitely and the same status message keeps getting logged without anyone re-checking the premise.

Source

Task-queue message history