One of the long held goals for the bot has been automated dependency handling, reducing maintenance burden to mostly troubleshooting. I expect there to be four main phases to implementing dependency handling, although each tool will have their own path through these phases.

Phases:

1. Audit all the feedstocks that are applicable. In this phase the tool is used to produce dependencies, ideally in meta.yaml format and write them into the graph database.
2. The dependencies need to be diffed against the existing dependencies for each recipe. This will be used to evaluate the suitability of the tool for generating dependencies. Note that this could be done for historical versions as well as current.
3. Provide hints in the PR body, this will help maintainers to implement changes as needed.
4. Once the system has been stable for a while and most of the changes have been correct the system can update the dependencies in the PR itself, provided that a needed flag is set in the conda-forge.yml configuration.