Laravel AI SDK 1.1 adds Agent Skills and Cohere chat
The minor release lets agents load instructions on demand via a new HasSkills contract, adds Cohere text generation, and extends tool approval to built-in tools.
Laravel AI SDK 1.1 shipped on 5 October, three months after v1.0, and the headline feature is Agent Skills: a new HasSkills contract gives an agent a LoadSkill tool so it can pull instructions from a folder only when a task actually needs them, rather than stuffing everything into the system prompt up front.
The release also adds Cohere as a text-generation provider through its v2 Chat API, with streaming, tool calls and structured output, where Cohere was previously embeddings-only in the SDK. Built-in tools — MCP, agent, similarity search and filesystem — can now require human approval before running, a capability that previously meant subclassing a tool yourself. Failover between providers is also more forgiving: if the next provider in the chain does not support a given provider tool, the SDK now drops that tool before the attempt instead of throwing.
class Support extends Agent
{
use HasSkills;
}
Why it matters: the SDK only reached v1.0 in September, so 1.1 is Laravel iterating fast on the rough edges production use turned up — skills keep large instruction sets out of every prompt, and approval on built-in tools closes a gap that previously forced custom code for anything touching the filesystem or an MCP server.