agents · review · first seen Oct 24, 01:28 UTC

CrewAI adds evaluation options in 1.15.29

CrewAI 1.15.29 shipped three evaluation changes: crewAI eval can run a crew for a verdict from one command, export evaluations as files, and keep full evaluation data in the project. A builder should care because the verdict and its underlying record can now be examined in one project workflow.

L0 · primary1 source3 claims checked4 min read

What shipped

The CrewAI 1.15.29 release notes say crewAI eval can run the crew itself for a verdict from one command. That is the concrete change: the evaluation command can initiate a crew run. The excerpt does not provide the command’s full syntax or explain how the verdict is calculated. I would treat the release entry as evidence that this option shipped, not as evidence that a particular verdict will be useful for my project. The supplied source is the GitHub release titled “1.15.29” in crewAIInc/crewAI. [1]

The same release says crewAI eval can export evaluations as files. This matters when a review needs an artifact beyond what appears during a command run. The entry confirms file export, but does not identify the format, destination, required option, or behavior when an export file already exists. I would not build an automated reader around an assumed structure. I would first export an evaluation and inspect the file produced by my installation. [1]

CrewAI 1.15.29 also keeps an evaluation’s full data in the project. That is distinct from file export: one change concerns data retained with the project, while the other concerns a file you can export. The release calls the retained data full, but does not enumerate its fields, describe its storage layout, or say what happens after later evaluations. I would inspect the project record before treating it as an audit trail. These are shipped entries in the release notes, not a field-level data specification. [1][2]

Why it matters if you build

If you already use CrewAI to build a crew, the practical workflow is to run that crew through crewAI eval, examine its verdict, and inspect the evaluation data left in the project. You can export an evaluation as a file when your review process needs a separate artifact. Those actions give you more to examine than a verdict alone. They do not establish that the verdict is correct or that the exported file contains every field retained in the project. The decision today is whether these artifacts help you review a real crew, not whether the release note proves evaluation quality. [1][2][3]

The work before adoption is validation. Choose a crew whose behavior you can inspect, then compare the verdict, project data, and exported evaluation against what your review actually needs. The cited entries give no runtime cost, model-usage cost, supported file types, or retention limits. They also do not say whether running the crew and exporting its evaluation happen in the same invocation. If your workflow depends on a particular format, cost ceiling, or repeatable verdict, the release excerpt is not enough to make that commitment. Check those points in your project before adding an automated gate. [1][2][3]

How to try it today

1. In a project using CrewAI 1.15.29, choose a crew whose output you can inspect.
2. Check the available help for crewAI eval, then use it to run the crew for a verdict.
3. Inspect the evaluation’s full data retained in the project and compare it with the verdict.
4. Check the available export option, then export an evaluation as a file.
5. Open the exported file and decide whether its contents meet your review needs.

What we checked

Every claim below was walked down the source ladder. The rung is how close the source is to the thing itself; the tense is the source's own.

1. shipped L0 · primary

   CrewAI 1.15.29 lets crewAI eval run the crew itself for a verdict from one command.

   per GitHub releases · crewAIInc/crewAI (L0)
2. shipped L0 · primary

   CrewAI 1.15.29 lets crewAI eval export evaluations as files.

   per GitHub releases · crewAIInc/crewAI (L0)
3. shipped L0 · primary

   CrewAI 1.15.29 keeps an evaluation's full data in the project.

   per GitHub releases · crewAIInc/crewAI (L0)

What is not confirmed

The release excerpt does not establish the verdict criteria, repeatability, exact command options, export format, export destination, retained fields, storage behavior, or operational costs. It does not show that the exported file and the project record contain identical data. Nor does it provide an example result to compare with a crew’s output. Those are specific questions to resolve in an actual installation before making this evaluation path a reporting or compliance dependency. [1][2][3]

Who should act, who should wait

Act now if you already have a CrewAI project and want to test whether a one-command crew run, retained evaluation data, and file export fit your review loop. Keep the first test small. Read the verdict and both forms of data rather than assuming their contents from the feature names. Wait before wiring the result into an automated gate or external report if you require defined verdict criteria, a stable file structure, known storage behavior, or a cost estimate. The 1.15.29 release confirms that the three options shipped; it does not settle those integration questions. [1][2][3]

Sources

• 1.15.29 — GitHub releases · crewAIInc/crewAI L0 · primary