diff --git a/README.md b/README.md
index 78b1c5f..454d8ea 100644
--- a/README.md
+++ b/README.md
@@ -34,4 +34,32 @@ Each space-separated text represents file in /extensions/Clip_IO/conditioning/ .
 For filename without extension, if it does not exist, will searched with appending ".csv" or ".pt" in that order.  
 If filename has space, you can enclose with single double-quotation or single single-quotation.  
 You can include prompt which to be processed by Clip, by enclosing prompt with triple double-quotation or triple single-quotation.  
-After gathering all conditionings, these conditionings will be concaterated.  
\ No newline at end of file
+After gathering all conditionings, these conditionings will be concaterated.  
+### Syntax for "Directive" mode
+**NOTE: Currently, because of syntax mess, "Directive" mode does not support Prompt editing and Alternating words.**
+In addition of "Simple" mode syntax, "Directive" mode supportes inline directives.  
+The syntax of inline directive is:
+?`DirectiveName`(`DirectiveInner`) or
+?`DirectiveName`_`DirectiveOrder`(`DirectiveInner`)
+"DirectiveName" is name of directive such as "eval" or "exec" (case-insensitive).  
+"DirectiveOrder" is order of directive.
+Larger order makes processing directive later.
+If directives with same order exists, these directives will be processed from left to right.
+If "DirectiveOrder" is absent, it will be treated as order is 0.
+#### Directives
+##### eval
+"eval" does component-wise python's eval to conditioning.
+Local objects for eval are:
+i: torch.Tensor : input conditioning
+o: torch.Tensor : output conditioning
+t: int : 0th dimension (token-wise) of index of input conditioning
+d: int : 1st dimension (dimension-wise) of index of input conditioning
+torch module and all objects in math module
+##### exec
+"exec" does component-wise python's exec.
+Local objects for exec are:
+i: torch.Tensor : input conditioning
+o: torch.Tensor : output conditioning
+t: int : 0th dimension (token-wise) of index of input conditioning
+d: int : 1st dimension (dimension-wise) of index of input conditioning
+torch module and all objects in math module
\ No newline at end of file