diff --git a/README.md b/README.md
index ede7499..8241ca3 100644
--- a/README.md
+++ b/README.md
@@ -1,3 +1,112 @@
-# EasyCTF 2016
+OpenCTF Problem Import via Github
+-------------------------------------------------
 
-Problems folder. Make a folder with your problem in it. Format coming soon... 
\ No newline at end of file
+Each problem belongs in its own folder under the root folder of the repository. The name of the folder will become a readable identifier for the problem, so it should only contain letters, numbers, and `-`.
+
+Inside this folder, there are 3 required files:
+
+- `problem.yml`: This is probably the most important file. It contains all the metadata for the problem, including:
+  - `title`: The title of the problem.
+  - `category`: The category of the problem.
+  - `value`: An integer value for the problem.
+  - `hint`: A hint for the problem. (no markdown supported)
+  - `autogen`: (`true`/`false`) whether or not the problem is autogenerated. If you put `true`, read on to see what needs to be added to the grader.
+- `description.md`: A problem description. As the extension suggests, you may use markdown in this file.
+- `grader.py`: A python file containing several functions that relate to grading the problem (and generating, if autogen is enabled).
+
+The Grader
+--------------
+
+At the most basic level, the grader should contain a `grade` function that takes in two parameters (you can call them whatever you like): (1) an instance of the random library used for autogenerating problems, and (2) the flag that the user actually typed into the input box. It should return a tuple `(correct, message)`, where `correct` is a boolean value signifying whether the user got the problem right, and `message`, a custom message to be displayed.
+
+An example of a basic grader looks like this:
+
+```python
+def grade(random, key):
+	if key.find("this_is_the_flag") != -1:
+		return True, "Correct!"
+	return False, "Nope."
+```
+
+If autogen is not enabled, please do not use `random` while judging the key. It's meant for checking problems that were autogenerated with the same seed. Also, note that I'm using `key.find` rather than comparing them with equals. This is useful if you want to match both `flag{this_is_the_flag}` and `this_is_the_flag`, or if you are trying to perform a case-insensitive match.
+
+If autogen were enabled, you would need to add an additional function to the grader, called `generate`. Essentially, this function is required for generating the problem. It takes in a single parameter, the same instance of `random` as the `grade` function gets, and must return a dictionary of values (obviously, this dictionary can be blank, but ideally it shouldn't be).
+
+Possible keys that you can include in your dictionary are:
+- `variables`: This should be another dictionary of variables that are inserted into the description during runtime. More details on how these variables are inserted comes in a later section. You would probably want to use the `random` variable that is passed to the function to generate a value that is different per team.
+- `files`: This is similar to `variables` in that it is a dictionary of keys that are inserted into the description. However, instead of returning variables directly, you should return a function that takes in the `random` variable and returns a File object (could be StringIO as well). Files generated this way are stored into the static container during runtime.
+
+An example using both `variables` and `files` is:
+
+```python
+from cStringIO import StringIO
+def gen_b(random):
+	b = random.choice(range(50, 100))
+	return b
+def generate(random):
+	a = random.choice(range(50, 100))
+	return dict(variables={
+		"a": a
+	}, files={
+		"b.txt": gen_b
+	})
+```
+
+Note that we are returning `gen_b` without `()`, since we want to return the function definition, rather than a value returned by the function. This way, the platform can call this function at runtime, generating a different problem for every user.
+
+**This part is very important**. Filenames usually have extensions, like `ciphertext.txt`. In this case, we are passing the filename into the description as a string. But Python template strings don't allow symbols such as periods. Therefore, ALL FUNCTION NAMES ARE SANITIZED by replacing anything matching this regex: `[^a-zA-Z]+` with `_` (underscore). This way, `ciphertext.txt` becomes `ciphertext_txt` in the description. Here is a description that uses the above generator.
+
+```markdown
+Help me decipher [this](${b_txt}).
+```
+
+Here is a full example using autogen:
+
+problem.yml:
+
+```yaml
+title: Caesar Cipher
+value: 20
+author: mzhang
+autogen: true
+```
+
+description.md:
+
+```markdown
+Help me decipher [this ciphertext](${ciphertext_txt}).
+```
+
+grader.py:
+
+```python
+from cStringIO import StringIO
+from string import maketrans
+
+flag = "caesar_cipher_is_fun!"
+alphabet = "abcdefghijklmnopqrstuvwxyz"
+
+def get_problem(random):
+	n = random.randint(1, 25)
+	salt = "".join([random.choice("0123456789abcdef") for i in range(6)])
+	return (n, salt)
+
+def generate_ciphertext(random):
+	n, salt = get_problem(random)
+	trans = maketrans(alphabet, alphabet[n:] + alphabet[:n])
+	ciphertext = ("easyctf{%s_%s}" % (flag, salt)).translate(trans)
+	return StringIO(ciphertext)
+
+def generate(random):
+	return dict(files={
+		"ciphertext.txt": generate_ciphertext
+	})
+
+def grade(random, key):
+	n, salt = get_problem(random)
+	if key.find("%s_%s" % (flag, salt)) >= 0:
+		return True, "Correct!"
+	return False, "Nope."
+```
+
+In this way, not only is the shift randomly generated, so is the actual flag itself. Note that I use a helper function, `get_random` to actually generate the numbers to ensure that the numbers generated are the same for both the generation and the grading.
\ No newline at end of file