agents · review · first seen Oct 29, 09:32 UTC

CrewAI releases 1.15.33 with skill search and flow inputs

CrewAI shipped 1.15.33 with organization skill search and a way to pass run inputs to a Python flow. If you manage shared skills or launch flows from the command line, those are the changes to test in your own workflow.

L0 · primary1 source4 claims checked4 min read

What shipped

The release adds crewai skill search to list an organization's skills. That is the stated scope: a listing command for organization skills. The release excerpt does not describe filtering, ranking, permission behavior, or how a returned skill is selected for a run. Treat this as a way to inspect a list, not as evidence that the command changes how an agent uses skills. [1]

CrewAI 1.15.33 passes crewai run --inputs to a Python flow. This is a specific change at the boundary between a command-line run and the flow it launches. The note does not specify an input schema, validation rules, or which flow step receives a value. If your run depends on supplied inputs, test that boundary with the inputs and flow you actually use rather than assuming the flag changes every kind of run. [1]

The same release adds deploy create --zip and prints the commit from which a zip is built. Those are two things to check during deployment: whether the zip option is available and whether the reported commit matches the source you meant to package. The note does not describe the zip contents or a deployment outcome. Separately, the release reports token usage for LiteLLM tool-call and structured-output calls. That is a reporting change, not a stated reduction in tokens used or a change to billing. [1][2]

Why it matters if you build

For a team that keeps skills at the organization level, the immediate workflow is easier to inspect: run the search command and examine the list it returns before deciding what to use. That matters most if your question is which skills are listed, not whether a particular skill will execute correctly. The release text gives no measure of search coverage or response quality, so a list from your own organization is the useful check. Do not treat discovery as proof of access, compatibility, or performance without testing those separately. [1]

For a builder launching a Python flow from the command line, input propagation is the change worth isolating. Try a run whose supplied values you can recognize in the flow's observed behavior, then compare that behavior with what you expected. A successful command invocation alone would not establish that the flow used every value as intended. The excerpt gives no limits on input size, accepted shapes, or error handling, so these remain questions for your test. This is a reason to check an existing path, not a basis for promising a faster or cheaper run. [1]

The two other changes affect operational checks rather than the core flow interface. If you package a deployment as a zip, the printed commit offers a source detail to record and compare with your intended source. It does not by itself verify the artifact's contents. If you monitor LiteLLM calls, token usage is now reported for tool-call and structured-output calls; check those figures against your own records before using them for cost decisions. Neither release note supplies prices, accuracy guarantees, or a before-and-after usage total. [1][2]

How to try it today

1. Use CrewAI 1.15.33 when testing these shipped changes.
2. Run crewai skill search and inspect the organization skills listed.
3. Run an existing Python flow with crewai run --inputs and check whether its behavior reflects the inputs you supplied.
4. If you use zip deployment, try deploy create --zip and note the printed source commit.
5. If you use LiteLLM tool calls or structured output, inspect the reported token usage for those calls.

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 1.15.33 adds crewai skill search to list an organization's skills.

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

   CrewAI 1.15.33 passes crewai run --inputs to a Python flow.

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

   CrewAI 1.15.33 adds deploy create --zip.

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

   CrewAI 1.15.33 reports token usage for LiteLLM tool-call and structured-output calls.

   per GitHub releases · crewAIInc/crewAI (L0)

What is not confirmed

The notes do not establish skill-search ranking or permissions, the accepted shape or validation of flow inputs, the contents or success of a zip deployment, or whether reported token usage matches a provider's bill. They also do not quantify time saved or cost changed. These are limits of the supplied release excerpt, not evidence that any of those behaviors fail. [1][2][3][4]

Who should act, who should wait

Act now if you already list organization skills, launch Python flows with supplied inputs, package zip deployments, or inspect LiteLLM token usage: each change has a narrow check you can run against your own setup. Wait on a broader adoption decision if you need documented search semantics, input validation, artifact guarantees, or billing reconciliation first. The release establishes that these changes shipped in 1.15.33; it does not establish those additional guarantees. [1][2][3][4]

Sources

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