gdharness 0.6.5 → 0.6.7

package/build/cli.js

@@ -15850,9 +15850,15 @@ var init_tool_definitions = __esm(() => {
           description: "click: the Control to click, at its centre, or the 3D node to click, where it is drawn. choose: the PopupMenu, or the OptionButton or MenuButton in front of one."
         },
         action: { type: "string", description: "action: the InputMap action name." },
-        pressed: { type: "boolean", description: "Press or release. Default true." },
+        pressed: {
+          type: "boolean",
+          description: "action, key: leave it out and the press is a whole one, down and up a frame apart. true holds it down, false lets go of one being held."
+        },
         strength: { type: "number", description: "action: 0 to 1. Default 1." },
-        keycode: { type: "string", description: 'key: the key name, such as "Space" or "A".' },
+        keycode: {
+          type: ["string", "number"],
+          description: 'key: the key name, such as "Space" or "A", or its Godot keycode.'
+        },
         text: {
           type: "string",
           description: "text: what to type. A newline is Enter and a tab is Tab. choose: the item to take, by what it says."
@@ -15867,9 +15873,8 @@ var init_tool_definitions = __esm(() => {
         x: { type: "number", description: "mouse_click, mouse_motion: window pixels." },
         y: { type: "number", description: "mouse_click, mouse_motion: window pixels." },
         button: {
-          type: "string",
-          enum: ["left", "right", "middle"],
-          description: "click, mouse_click: default left."
+          type: ["string", "number"],
+          description: "click, mouse_click: left, right, middle, wheel_up or wheel_down, or a button number. Default left."
         },
         doubleClick: { type: "boolean", description: "click, mouse_click: default false." },
         relativeX: { type: "number", description: "mouse_motion: movement since the last event." },
@@ -15885,8 +15890,8 @@ var init_tool_definitions = __esm(() => {
           summary: "take an item out of a menu, by what it says or by where it is in the list. A menu's items are drawn rather than built, so there is nothing to click: the item takes the focus and Enter presses it, which is the engine's own path and needs no window. Answers with what was chosen and what the button in front of it shows now",
           requires: ["nodePath"]
         },
-        action: { summary: "press or release an action", requires: ["action"] },
-        key: { summary: "press or release a key", requires: ["keycode"] },
+        action: { summary: "press an action, or hold it", requires: ["action"] },
+        key: { summary: "press a key, or hold it", requires: ["keycode"] },
         text: {
           summary: "type a string wherever the focus is, a character at a time",
           requires: ["text"]
@@ -35855,6 +35860,45 @@ import { promisify as promisify5 } from "node:util";
 function patienceForFrames(frames, atLeast) {
   return Math.max(atLeast, frames / SLOWEST_FRAME_RATE * 1000 + atLeast);
 }
+function isOfType(value, kind) {
+  switch (kind) {
+    case "string":
+      return typeof value === "string";
+    case "number":
+      return typeof value === "number" && Number.isFinite(value);
+    case "integer":
+      return Number.isInteger(value);
+    case "boolean":
+      return typeof value === "boolean";
+    case "array":
+      return Array.isArray(value);
+    case "object":
+      return typeof value === "object" && value !== null && !Array.isArray(value);
+    default:
+      return true;
+  }
+}
+function describeValue(value) {
+  if (Array.isArray(value)) {
+    return "a list";
+  }
+  return VALUE_NAMES[typeof value] ?? typeof value;
+}
+function wrongTypes(spec, args) {
+  const complaints = [];
+  for (const [name, value] of Object.entries(args)) {
+    if (value === undefined || value === null) {
+      continue;
+    }
+    const declared = spec.parameters[name]?.["type"];
+    const wanted = typeof declared === "string" ? [declared] : Array.isArray(declared) ? declared : [];
+    if (wanted.length === 0 || wanted.some((kind) => isOfType(value, kind))) {
+      continue;
+    }
+    complaints.push(`${name} as ${wanted.join(" or ")}, not ${describeValue(value)}`);
+  }
+  return complaints;
+}
 function escapingPropertyValue(properties) {
   if (typeof properties !== "object" || properties === null) {
     return;
@@ -36193,6 +36237,10 @@ class GodotServer {
         response: this.createErrorResponse(`${spec.name} does not take ${unknown.join(", ")}. It takes: ${[...known].join(", ")}.`)
       };
     }
+    const wrong = wrongTypes(spec, args);
+    if (wrong.length > 0) {
+      return { ok: false, response: this.createErrorResponse(`${spec.name} takes ${wrong.join("; ")}.`) };
+    }
     let op = null;
     if (spec.operations) {
       const valid = Object.keys(spec.operations);
@@ -37411,7 +37459,7 @@ async function runGodotServer() {
   const server = new GodotServer;
   await server.run();
 }
-var UPDATE_NOTICE_EVERY = 500, FEEDBACK_NOTICE_EVERY = 250, run5, __dirname2, EDITOR_RESTART_TIMEOUT_MS = 90000, SLOWEST_FRAME_RATE = 20, BRIDGE_RETRY_MS = 2000, PATH_SOLUTIONS, PROJECT_FILE_ARGUMENTS, DEBUG_STATE_CALLS, HEADLESS_OPERATIONS, PROJECT_INFO_SECTIONS;
+var UPDATE_NOTICE_EVERY = 500, FEEDBACK_NOTICE_EVERY = 250, run5, __dirname2, EDITOR_RESTART_TIMEOUT_MS = 90000, SLOWEST_FRAME_RATE = 20, VALUE_NAMES, BRIDGE_RETRY_MS = 2000, PATH_SOLUTIONS, PROJECT_FILE_ARGUMENTS, DEBUG_STATE_CALLS, HEADLESS_OPERATIONS, PROJECT_INFO_SECTIONS;
 var init_server3 = __esm(() => {
   init_mcp();
   init_stdio2();
@@ -37438,6 +37486,12 @@ var init_server3 = __esm(() => {
   init_update_check();
   run5 = promisify5(execFile5);
   __dirname2 = dirname9(fileURLToPath4(import.meta.url));
+  VALUE_NAMES = {
+    string: "a string",
+    number: "a number",
+    boolean: "a boolean",
+    object: "an object"
+  };
   PATH_SOLUTIONS = [
     'Give the path relative to the project, such as "scenes/main.tscn" or "res://scenes/main.tscn"',
     "Point projectPath at the project the file belongs to"
@@ -38680,6 +38734,15 @@ import { basename, dirname as dirname4, join as join9 } from "node:path";
 
 // src/tool-reference.ts
 init_tool_definitions();
+function namedType(declared) {
+  if (typeof declared === "string") {
+    return declared;
+  }
+  if (Array.isArray(declared) && declared.every((kind) => typeof kind === "string")) {
+    return declared.join(" or ");
+  }
+  return "any";
+}
 function renderToolsMarkdown() {
   const lines = [
     "# Tools",
@@ -38710,7 +38773,7 @@ function renderToolsMarkdown() {
     if (parameters.length > 0) {
       lines.push("Arguments:", "");
       for (const [name, schema] of parameters) {
-        const type = typeof schema["type"] === "string" ? schema["type"] : "any";
+        const type = namedType(schema["type"]);
         const note = typeof schema["description"] === "string" ? ` ${schema["description"]}` : "";
         lines.push(`- \`${name}\` (${type}):${note}`);
       }
@@ -38767,6 +38830,7 @@ that is running, and the project on disk. ${TOOL_SPECS.length} tools, named \`do
 | Fill in a field | \`runtime_input click\` on it, then \`runtime_input text\` |
 | Click somebody standing in a 3D room | \`runtime_input click\` on the Node3D, which aims at what it draws |
 | Pick something out of a dropdown | \`runtime_input choose\` on it, by what the item says |
+| Answer a dialog, or press a bound key | \`runtime_input action\` or \`key\`, a whole press unless you hold it |
 | Where a 3D node is on screen | \`runtime_inspect\` \`rect\`, rather than unprojecting by hand |
 | Wait for something | \`runtime_wait\`, never a sleep |
 | A picture, for a person who asked to see one | \`runtime_capture\` |

package/build/godot/addons/gdharness_runtime/runtime_input.gd

@@ -30,9 +30,19 @@ func _init(host: Node, values: Values) -> void:
 	_values = values
 
 
+## Presses [param params].action, and lets go of it again unless asked to hold it.
+##
+## The whole press by default, for the reason [method click] is the whole click: an action left
+## down is not a press that did nothing, it is a press that never ends, and everything reading
+## [method Input.is_action_pressed] goes on seeing it for the rest of the session. Answering a
+## dialog with `ui_accept` took two calls and left the first one held, which is the shape that
+## found this.
+##
+## `pressed` is how a caller says otherwise: true holds it down, false lets go of one being held.
 func inject_action(params: Dictionary) -> Dictionary:
 	var action: String = String(params.get("action", ""))
-	var pressed: bool = bool(params.get("pressed", true))
+	var held: bool = bool(params.get("pressed", true))
+	var whole: bool = not params.has("pressed")
 	var strength: float = float(params.get("strength", 1.0))
 
 	if action.is_empty():
@@ -41,18 +51,36 @@ func inject_action(params: Dictionary) -> Dictionary:
 	if not InputMap.has_action(action):
 		return {"type": "error", "message": "Action not found: " + action}
 
+	_say_action(action, whole or held, strength)
+	if whole:
+		# A frame between the halves, as a click has: a listener that acts on the press and a
+		# listener that acts on the release both get their own frame to do it in.
+		await _host.get_tree().process_frame
+		_say_action(action, false, strength)
+		await _host.get_tree().process_frame
+
+	return {
+		"type": "input_injected",
+		"input_type": "action",
+		"action": action,
+		"pressed": held and not whole,
+		"whole": whole,
+	}
+
+
+func _say_action(action: String, down: bool, strength: float) -> void:
 	var event: InputEventAction = InputEventAction.new()
 	event.action = action
-	event.pressed = pressed
+	event.pressed = down
 	event.strength = strength
 	Input.parse_input_event(event)
 
-	return {"type": "input_injected", "input_type": "action", "action": action, "pressed": pressed}
-
 
+## Presses one key, and lets go of it again unless asked to hold it. See [method inject_action].
 func inject_key(params: Dictionary) -> Dictionary:
 	var keycode_raw: Variant = params.get("keycode", 0)
-	var pressed: bool = bool(params.get("pressed", true))
+	var held: bool = bool(params.get("pressed", true))
+	var whole: bool = not params.has("pressed")
 	var key_label: String = String(params.get("key_label", ""))
 
 	if keycode_raw is String:
@@ -62,7 +90,7 @@ func inject_key(params: Dictionary) -> Dictionary:
 	var keycode: int = 0 if keycode_raw is String else int(keycode_raw)
 
 	var event: InputEventKey = InputEventKey.new()
-	event.pressed = pressed
+	event.pressed = whole or held
 
 	if not key_label.is_empty():
 		event.keycode = OS.find_keycode_from_string(key_label)
@@ -92,6 +120,12 @@ func inject_key(params: Dictionary) -> Dictionary:
 	event.unicode = _glyph_of(event.keycode, event.shift_pressed)
 
 	Input.parse_input_event(event)
+	if whole:
+		await _host.get_tree().process_frame
+		var up: InputEventKey = event.duplicate()
+		up.pressed = false
+		Input.parse_input_event(up)
+		await _host.get_tree().process_frame
 
 	return {
 		"type": "input_injected",
@@ -102,7 +136,8 @@ func inject_key(params: Dictionary) -> Dictionary:
 		"ctrl": event.ctrl_pressed,
 		"alt": event.alt_pressed,
 		"unicode": event.unicode,
-		"pressed": pressed
+		"pressed": held and not whole,
+		"whole": whole,
 	}
 
 
@@ -190,6 +225,8 @@ func inject_mouse_click(params: Dictionary) -> Dictionary:
 		return {"type": "error", "message": point}
 	var position: Vector2 = point
 	var button: int = _resolve_mouse_button(params.get("button", MOUSE_BUTTON_LEFT))
+	if button < 0:
+		return _no_such_button(params.get("button"))
 	var pressed: bool = bool(params.get("pressed", true))
 	var double: bool = bool(params.get("doubleClick", false))
 
@@ -262,6 +299,24 @@ static func _centre_of(control: Control) -> Vector2:
 ## the control within the outer one, so the outer has to be asked after the inner has finished
 ## moving it. A control with no ScrollContainer over it moves nothing and answers false, which is
 ## what keeps the refusal below saying the right thing about a control that is simply off screen.
+## Whether [param centre] is somewhere a click can reach it: inside the viewport, and inside every
+## ScrollContainer between the control and the root, each of which clips what it holds.
+static func _in_sight(control: Control, viewport: Viewport, centre: Vector2) -> bool:
+	if not viewport.get_visible_rect().has_point(centre):
+		return false
+	var walking: Node = control.get_parent()
+	while walking != null:
+		var holder: ScrollContainer = walking as ScrollContainer
+		# Carried into the same space the centre is in, which is the canvas rather than the
+		# container's own: the two are only the same while nothing above it is transformed.
+		if holder != null:
+			var seen: Rect2 = holder.get_global_transform_with_canvas() * Rect2(Vector2.ZERO, holder.size)
+			if not seen.has_point(centre):
+				return false
+		walking = walking.get_parent()
+	return true
+
+
 static func _scroll_into_view(control: Control) -> bool:
 	var moved: bool = false
 	var walking: Node = control.get_parent()
@@ -304,12 +359,17 @@ func click(params: Dictionary) -> Dictionary:
 	var viewport: Viewport = _clicking_viewport(control)
 	var centre: Vector2 = _centre_of(control)
 
-	# A control below the fold of a ScrollContainer is not out of reach, it is one scroll away,
+	# A control out of sight inside a ScrollContainer is not out of reach, it is one scroll away,
 	# which is what a person does without thinking about it before they click. Refusing it
 	# instead sent callers to emit the button's own signal, which presses nothing, runs none of
 	# the input path and reports success.
+	#
+	# Out of sight against what it is clipped to rather than against the viewport: a row scrolled
+	# off the top of its container is inside the viewport, behind whatever is drawn up there, so
+	# the click went to that instead and said so. Measured on a hall whose staff panel had been
+	# scrolled past: the button was at y 69 and its container started at y 166.
 	var scrolled: bool = false
-	if not viewport.get_visible_rect().has_point(centre):
+	if not _in_sight(control, viewport, centre):
 		scrolled = _scroll_into_view(control)
 		if scrolled:
 			# A container moves its child on the next layout pass rather than inside the call.
@@ -334,6 +394,8 @@ func click(params: Dictionary) -> Dictionary:
 			)
 		}
 	var button: int = _resolve_mouse_button(params.get("button", MOUSE_BUTTON_LEFT))
+	if button < 0:
+		return _no_such_button(params.get("button"))
 	var double: bool = bool(params.get("double", false))
 
 	# Pushed into the viewport rather than through Input: Input accumulates events and flushes
@@ -594,6 +656,8 @@ func _click_in_the_world(node_path: String, item: Node3D, params: Dictionary) ->
 
 	var position: Vector2 = viewport.get_final_transform() * aim
 	var button: int = _resolve_mouse_button(params.get("button", MOUSE_BUTTON_LEFT))
+	if button < 0:
+		return _no_such_button(params.get("button"))
 	var double: bool = bool(params.get("double", false))
 
 	viewport.push_input(_motion(position, Vector2.ZERO))
@@ -642,6 +706,23 @@ func _button(position: Vector2, button: int, pressed: bool, double: bool) -> Inp
 	return event
 
 
+## What to say about a name that is not a mouse button, with the ones that are.
+func _no_such_button(named: Variant) -> Dictionary:
+	return {
+		"type": "error",
+		"message":
+		(
+			"%s is not a mouse button. It takes: left, right, middle, wheel_up, wheel_down, or the number of one."
+			% JSON.stringify(named)
+		)
+	}
+
+
+## The button [param raw] names, or -1 for a name that is not one of them.
+##
+## A name nobody recognises used to come back as the left button, so a right click asked for by a
+## spelling this does not know went to the left button and reported the left button. The caller
+## reads the answer and believes it, which is the whole reason nothing here guesses.
 func _resolve_mouse_button(raw: Variant) -> int:
 	if raw is String:
 		var named: String = raw
@@ -657,5 +738,5 @@ func _resolve_mouse_button(raw: Variant) -> int:
 			"wheel_down", "wheeldown":
 				return MOUSE_BUTTON_WHEEL_DOWN
 			_:
-				return MOUSE_BUTTON_LEFT
+				return -1
 	return int(raw)

package/build/index.js

@@ -28404,9 +28404,15 @@ var TOOL_SPECS = [
         description: "click: the Control to click, at its centre, or the 3D node to click, where it is drawn. choose: the PopupMenu, or the OptionButton or MenuButton in front of one."
       },
       action: { type: "string", description: "action: the InputMap action name." },
-      pressed: { type: "boolean", description: "Press or release. Default true." },
+      pressed: {
+        type: "boolean",
+        description: "action, key: leave it out and the press is a whole one, down and up a frame apart. true holds it down, false lets go of one being held."
+      },
       strength: { type: "number", description: "action: 0 to 1. Default 1." },
-      keycode: { type: "string", description: 'key: the key name, such as "Space" or "A".' },
+      keycode: {
+        type: ["string", "number"],
+        description: 'key: the key name, such as "Space" or "A", or its Godot keycode.'
+      },
       text: {
         type: "string",
         description: "text: what to type. A newline is Enter and a tab is Tab. choose: the item to take, by what it says."
@@ -28421,9 +28427,8 @@ var TOOL_SPECS = [
       x: { type: "number", description: "mouse_click, mouse_motion: window pixels." },
       y: { type: "number", description: "mouse_click, mouse_motion: window pixels." },
       button: {
-        type: "string",
-        enum: ["left", "right", "middle"],
-        description: "click, mouse_click: default left."
+        type: ["string", "number"],
+        description: "click, mouse_click: left, right, middle, wheel_up or wheel_down, or a button number. Default left."
       },
       doubleClick: { type: "boolean", description: "click, mouse_click: default false." },
       relativeX: { type: "number", description: "mouse_motion: movement since the last event." },
@@ -28439,8 +28444,8 @@ var TOOL_SPECS = [
         summary: "take an item out of a menu, by what it says or by where it is in the list. A menu's items are drawn rather than built, so there is nothing to click: the item takes the focus and Enter presses it, which is the engine's own path and needs no window. Answers with what was chosen and what the button in front of it shows now",
         requires: ["nodePath"]
       },
-      action: { summary: "press or release an action", requires: ["action"] },
-      key: { summary: "press or release a key", requires: ["keycode"] },
+      action: { summary: "press an action, or hold it", requires: ["action"] },
+      key: { summary: "press a key, or hold it", requires: ["keycode"] },
       text: {
         summary: "type a string wherever the focus is, a character at a time",
         requires: ["text"]
@@ -28575,6 +28580,51 @@ var SLOWEST_FRAME_RATE = 20;
 function patienceForFrames(frames, atLeast) {
   return Math.max(atLeast, frames / SLOWEST_FRAME_RATE * 1000 + atLeast);
 }
+function isOfType(value, kind) {
+  switch (kind) {
+    case "string":
+      return typeof value === "string";
+    case "number":
+      return typeof value === "number" && Number.isFinite(value);
+    case "integer":
+      return Number.isInteger(value);
+    case "boolean":
+      return typeof value === "boolean";
+    case "array":
+      return Array.isArray(value);
+    case "object":
+      return typeof value === "object" && value !== null && !Array.isArray(value);
+    default:
+      return true;
+  }
+}
+var VALUE_NAMES = {
+  string: "a string",
+  number: "a number",
+  boolean: "a boolean",
+  object: "an object"
+};
+function describeValue(value) {
+  if (Array.isArray(value)) {
+    return "a list";
+  }
+  return VALUE_NAMES[typeof value] ?? typeof value;
+}
+function wrongTypes(spec, args) {
+  const complaints = [];
+  for (const [name, value] of Object.entries(args)) {
+    if (value === undefined || value === null) {
+      continue;
+    }
+    const declared = spec.parameters[name]?.["type"];
+    const wanted = typeof declared === "string" ? [declared] : Array.isArray(declared) ? declared : [];
+    if (wanted.length === 0 || wanted.some((kind) => isOfType(value, kind))) {
+      continue;
+    }
+    complaints.push(`${name} as ${wanted.join(" or ")}, not ${describeValue(value)}`);
+  }
+  return complaints;
+}
 var BRIDGE_RETRY_MS = 2000;
 var PATH_SOLUTIONS = [
   'Give the path relative to the project, such as "scenes/main.tscn" or "res://scenes/main.tscn"',
@@ -28967,6 +29017,10 @@ class GodotServer {
         response: this.createErrorResponse(`${spec.name} does not take ${unknown.join(", ")}. It takes: ${[...known].join(", ")}.`)
       };
     }
+    const wrong = wrongTypes(spec, args);
+    if (wrong.length > 0) {
+      return { ok: false, response: this.createErrorResponse(`${spec.name} takes ${wrong.join("; ")}.`) };
+    }
     let op = null;
     if (spec.operations) {
       const valid = Object.keys(spec.operations);

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "gdharness",
-  "version": "0.6.5",
+  "version": "0.6.7",
   "mcpName": "io.github.Aureliolo/gdharness",
   "description": "A harness for driving a Godot 4 project from an agent: editor addons, a runtime bridge into the running game, an MCP server in front of them, and a CLI that installs and diagnoses the Godot side.",
   "type": "module",