had this happen with a retry loop. hit $80 on anthropic before i caught it. how does this handle retries? seems like an agent could just keep retrying and blow past the limit
That is exactly why we enforce a hard budget cap (default is $1.00).
The system runs a pre-flight check before every LLM call. If you're about to go over budget, it kills the process immediately (SentinelBudgetExceeded).
We don't have a specific "max retries" counter wired up yet, but I'll likely add that soon based on your feedback. For now, the budget cap would have caught it at $1.
I just added loop detection based on your comment. It fingerprints tool actions (tool name + args) and kills the agent if it tries the exact same thing 5 times in a row.
had this happen with a retry loop. hit $80 on anthropic before i caught it. how does this handle retries? seems like an agent could just keep retrying and blow past the limit
That is exactly why we enforce a hard budget cap (default is $1.00).
The system runs a pre-flight check before every LLM call. If you're about to go over budget, it kills the process immediately (SentinelBudgetExceeded).
We don't have a specific "max retries" counter wired up yet, but I'll likely add that soon based on your feedback. For now, the budget cap would have caught it at $1.
I just added loop detection based on your comment. It fingerprints tool actions (tool name + args) and kills the agent if it tries the exact same thing 5 times in a row.