FOR /F "tokens=*" %%G IN ('dir /B /S /A-D') DO whatever.exe "%%G"
Where the /S option means recursive, and /A-D makes it not pass dir names to whatever.exe.

I usually locate/reverse engineer the model loaders using IDA and then use PIX where possible to make sense of data, like you can tell the input layout of vertex buffers and such.

Or use http://msdn.microsoft.com/en-us/library/a3140177.aspx which results in far less instructions per swap.

What's the format of your filenames? Are they similar to 3ad1623850608243.0.bd007e20bc2b61af?

The audioformat for the files I checked is ADPCM with 64 samples per block (they use wwise for audio), so it would be pretty simple to write a tool to convert them to PCM.

09E2DB206ACBCB09 is the hash for "stream", which is their sound file extension.

Here's a fixed ext_info.txt: http://sktest.aruarose.com/ext_info.txt.

Files inside the bundles are found by two hashes each: one for the extension, another for path + name. So if you get filenames containing the extension hash after extracting, you can just batch rename them.

Can you provide one of the normal maps?

Well, provide an entire file and I can look.

Looks like CE3, yeah. How does one get this game?

Here's the source for my tool. When it says "Failed to decompress" and still produces a valid .bmp, it means the input file was too large.
http://sktest.aruarose.com/BmpDecompress.7z

Could be LZSS or something similar. Try to figure out what this is from!

Edit: yeah, it's LZSS. Not perfect, but managed to decompress it some:


Edit 2: fixed

The var names are from the Source engine send tables. And the rest is possibly something like asserts or some other kind of debugging aid.

Files in the archives seem to be looked up with just the index.

Compared to like the decompressor found here: http://my.execpc.com/~geezer/code/lzss.c , the only changes are the length being stored in the upper four bits/the rest of the offset being in the lower four bits and the -1 on the position when reading from the buffer. So it might be easy to take the co...

It's just LZSS. I made a tool to decompress the files.
http://sktest.aruarose.com/TM2Decompress.7z

The archive also contains a fixed File00000000, as the provided one was too large.