It is easiest to generate tools for galaxy when the applications or scripts can take arbitrarily named input files and generate output to given path names.  
Input directories, output directories are very convenient on the command line, but more of a challenge when crafting a galaxy tool. 
That said, many applications require a wrapper script to work with in galaxy.   
Thank you for the consistent script_info[] help/usage syntax in the qiime scripts,  which enabled me to generate a skeleton galaxy tool_config file for each qiime script.

I had some time last spring to work on integrating qiime into galaxy.
Unfortunately, I haven't had any time since to work on this.
I put those partial results  on the Galaxy Tool Shed:    http://toolshed.g2.bx.psu.edu/
There's a continuing effort at George Mason University to incorporate qiime into galaxy tools, so you may want to ask them what they need. 


I started by generating galaxy tool_config files, e.g. align_seqs.xml,  by using python to get the script_info[] from the qiime script:

$ cat generate_tool_config.bash
#!/usr/bin/env bash
python $1 > ${1%.*}.help
cat tool_template.txt | sed "s/__TOOL_BINARY__/${1}/" | python -i $1 -h > ${1%.*}.log

(I'll attach tool_template.txt )

This generated skeleton tool_config .xml files that I could then edit as needed.
( http://wiki.g2.bx.psu.edu/Admin/Tools/Tool%20Config%20Syntax )

I originally was calling all qiime scripts from a tool wrapper:  qiime_wrapper.py
But, if a script can be called with any input filepaths and write its results to any filepaths, and only writes to STDERR when it fails, then you could call that script directly.
  

When should you use a tool_wrapper or call the qiime script directly?
  Many of the qiime scripts could probably be called directly, especially if it can be called with arbitary input/output file pathnames.
  The reasons for using a tool wrapper may be if input/output needs to be manipulated, moved, renamed in order to be used by the qiime script.
  You'll also need a tool wrapper if the names or number of the output files can not be determined from the parameter settings.
  ( http://wiki.g2.bx.psu.edu/Admin/Tools/Multiple%20Output%20Files )
  If your tool relies on a file ext to determine a format, you'll have to rename the input.
  ( Galaxy dataset pathnames will look something like:  /<your_galaxy_file_path>/072/dataset_72931.dat )
  The format/type of a dataset is stored in its metadata, so the tool_config can use that information, especially if a script can take muliple alternative input formats.
  A tool_wrapper can also be used to manage the stdout or stderr from a tool.   Galaxy currently interprets any output on stderr as a failure.



A couple changes in galaxy should make somethings easier than when I first attempted this:  
  - galaxy now accepts dataset requests with sub directories. ( https://bitbucket.org/galaxy/galaxy-central/issue/494/support-sub-dirs-in-extra_files_path-patch )
    That means that output HTML files with links into sub directories can be left intact, with the html copied to the output dataset and the linked files to its "extra_files_path".
  - if you know the pathname of an output relative to the working directory, galaxy can copy it automatically to the output dataset using the from_work_dir attribute.
    ( see example in:  https://bitbucket.org/galaxy/galaxy-central/src/21b645303c02/tools/ngs_rna/tophat_wrapper.xml )

Datatypes
  You may want to create new datatypes to make it easier for the user to correctly select inputs to a tool from previous outputs. 
  For example, the qiime mapping file is a tabular file with specific requirements.  I put a 'qiimemapping' datatype in lib/galaxy/datatypes/metagenomics.py and datatypes_conf.xml
  so an input could generate a select list containing only qiimemapping datasets rather than all tabular ones. 

Generating a configfile
  You can generate configfiles in the galaxy tool_config .xml file.   The configfile is generated by the Cheetah interpreter just as the commandline is.
  see:  alpha_rarefaction.xml

The qiime_wrapper.py was patterned after the mothur_wrapper.py   with some of the same wrapper params to handle run time determined output (perhaps not needed):
  --galaxy_datasets
         a comma separated list of regex:output_dataset the wrapper searches the working_dir and copies the file that matches the regex to the outout dataset
         if the exact pathname is known, use the "from_work_dir" attribute instead
  --galaxy_datasetid
         would be an output dataset id that would be used to dynamically create additional new datasets at job termination
         ( http://wiki.g2.bx.psu.edu/Admin/Tools/Multiple%20Output%20Files "Number of Output datasets cannot be determined until tool run")
  --galaxy_new_datasets
         a comma separated list of regex:datatype used to dynamically create additional new datasets at job termination
  --galaxy_new_files_path
         the galaxy dir for dynamically generated output datasets