tools/

Any @AITool-annotated function under tools/ becomes a callable tool.

tools/

Any .bx file under tools/ (searched recursively, so subfolders like tools/nested/Search.bx work too) that declares an @AITool-annotated function becomes a callable tool for the agent.

// tools/Greeter.bx
class {

	@AITool( "Say hello to someone by name." )
	function sayHello( name ) {
		return "Hello, " & arguments.name & "!";
	}

}

BxAgents doesn't scan or interpret @AITool annotations itself - it discovers one entry per .bx file (for name-collision checking) and then copies the whole tools/ folder verbatim into the generated app. At startup, the generated app calls bx-ai's own scanner:

aiToolRegistry().scan( "tools" )

which does the real reflection work (walking the copied files, instantiating each class, finding @AITool functions). This statement is only emitted at all if the project actually has a tools/ folder with files in it.

Naming

The discovered entry name is the file's own base name (Greeter.bx → Greeter). Two tool files with the same base name (even in different subfolders, since discovery is flat by base name) fail validation with a duplicate-name error.

What's excluded

• Dotfiles (anything starting with .) are ignored by discovery.
• .env/dotfiles inside tools/ are never copied into the build output, even if present - the copy step explicitly excludes them.

Rebuilds

tools/ is copied with a wipe-then-write strategy: the generated app's own tools/ directory is deleted and rewritten on every build, so a file removed from your project's tools/ folder doesn't linger as stale output from a previous build.