CrewAI 1.15.9 stops reporting tool failures as successes

CrewAI's July 30 release fixed a bug that reported failed tool calls as successful, per the project's GitHub changelog.

That's a bigger deal than a routine patch note. If a tool call failed and the framework logged it as a win anyway, the crew's next reasoning step built on a false premise. Nothing in the trace explained why the output was off.

Version 1.15.9 also adds a FlowFailedEvent, so flow-level failures now generate a signal too. Before this release, a crew could fail at the flow level and leave no distinct failure event behind, per the release notes.

The fix is straightforward. The harder problem is everything logged before it. If you run CrewAI in production, some traces marked "success" may actually record a tool that broke, with the agent pushing forward anyway.

I'd bet most teams don't go back and re-audit old runs after a bug fix like this. That means whatever weird outputs got shrugged off as "the model doing something odd" over the past several months are worth a second look. If your CrewAI logs show a clean success next to an output that never made sense, this bug is a strong candidate for why.