infer action type from instruction (#1231)

skyvern/forge/prompts/skyvern/infer-action-type.j2

@@ -0,0 +1,16 @@
+You are a browser agent performing actions on the web. You are instructed to take a single action. Help to identify which action type should be taken according to the action instruction.
+
+MAKE SURE YOU OUTPUT VALID JSON. No text before or after JSON, no trailing commas, no comments (//), no unnecessary quotes, etc.
+
+Reply in the following JSON format:
+{
+    "thought": str, // A string to describe how to infer the action type from the action instruction.
+    "confidence_float": float, // The confidence of the action. Pick a number between 0.0 and 1.0. 0.0 means no confidence, 1.0 means full confidence
+    "action_type": str, // It's a string enum: "CLICK", "INPUT_TEXT", "UPLOAD_FILE", "SELECT_OPTION". "CLICK" means user wants to click. "INPUT_TEXT" means user wants to input text. "UPLOAD_FILE" means user wants to upload a file. "SELECT_OPTION" means user wants to select an option.
+    "error": str, // It's a string enum to describe error. Null if you can identify the action as one of the defined action type. Use "UNKNOWN_ACTION" if none of the defined action type matched. Use "MULTIPLE_ACTIONS" if the instruction includes multiple actions.
+}
+
+Action instruction
+```
+{{ navigation_goal }}
+```