Laravel Installer v5.27.0
adds support for AI agent environments. When
laravel new
detects it's running inside an agent, it suppresses interactive prompts and decorative output and prints a single line of structured JSON to stdout instead.
The change comes from 乔·坦南鲍姆 , and it covers the two things an agent actually needs to know: did it work, and if not, why?
A successful run produces:
{
“成功”
:
真的
,
“姓名”
:
“我的应用”
,
"directory"
:
"/Users/you/code/my-app"}On failure, the output includes the error message, a path to the full installer log, and a tail of that log — enough context for an agent to understand what went wrong without having to chase down a temp file:
{
“成功”
:
错误的
,
“姓名”
:
“我的应用”
,
"directory"
:
"/Users/you/code/my-app"
,
“错误”
:
"Composer installation failed."
,
“日志”
:
"/tmp/laravel-installer-abc123"
,
"tail"
:
“……”}Detection happens automatically through a new
Agent
class that checks for environment variables set by common AI coding tools, including Claude Code, Cursor, GitHub Copilot, and Windsurf. No flags to pass or configuration needed — the installer figures out the context on its own.
For agentic workflows the agent runs the command, reads the JSON result, and either continues or handles the error — without trying to parse terminal output that wasn't meant for it.
