A moderation bot for Reddit, based on AutoModerator.
AutoModerator already exists, why reinvent the wheel?
The short answer: this wheel is a roughly hewn stone. It may be a wheel, but it could definitely be better.
The long answer...
In AutoModerator, configuration comprises a collection of Rules. Each Rule is itself a mixed collection of Match Conditions and response Actions. The match conditions specify criteria for selecting posts and comments, which are then responded to by the actions. For example:
---
# these are the conditions:
title (contains): ["Some", "shitpost", "phrases"]
kind: 'post' # redundant, but whatever
reports: 2 # at least two reports
# these are the responses:
action: remove
comment: 'No shitposts'
---
Besides convention, there is no clear distinction between the matching conditions and the responses. There's also no way to take the same set of responses to a variety of conditions except by copy-paste duplication of the responses section. These rules are highly constrained, as they must select the thing (the Reddit codebase calls posts, comments, and all other data points a thing) itself that you want to act on.
Some kinds of contextual information can be selected by using "sub-groups", an ungainly solution which is still highly limited. For example, you can check some things about the user account associated with the matched thing, as well as checking some things about the parent submission of matched comment items. See the example at the link, it's just atrocious:
---
# this example sets sumbmission flair if a known/approved user calls it a repost
type: comment
body: "repost"
is_top_level: true
author:
# this is a matching condition
flair_css_class: "trusted"
parent_submission:
# but this is a reaction
set_flair: "Possible Repost"s
---
I wanted a better way: to cleanly and clearly separate the matching conditions from the responses, to be able to group multiple sets of different matching criteria and have the same response, and to be able to target arbitrary contextual information for matching and response. For example, we can set different report thresholds for comments and posts:
---
criteria:
- type: post
reports: 5
- type: comment
reports: 2
actions:
action: remove
---
Or we should be able to remove both comments and submissions based on reply by trusted user, locking and sticky-ing the comment if it's a submission:
---
criteria:
type: comment
body (regex): '^!remove\s(.+)$'
is_reply: true
author:
contributor: true # let's make the approved submitters list useful
actions:
parent:
action: remove
lock: true
reply (sticky): "This {{parent.type}} has been removed: {{match-1}}"
---
This should give a flavor for the eventual capabilities of this bot, and its clear advantages over AutoModerator. Note that the examples are probably not yet functional.
Python: 3.6.3 Others: see requirements.txt
Top level dependencies are recorded in requirements.txt. An exact list of
operational state dependencies is recorded in requirements.deploy.txt. The
former is used during development to maintain the actual dependencies of the
project. The latter is used for deployment to ensure exact version
compatibility. Based on Kenneth Jones'
workflow.
During initial project setup, use pip install -r requirements.depoly.txt to ensure compatibility with known functional
versions.
During development, keep dependencies up to date:
-
Add or update entries in
requirements.txt -
Get the latest versions:
pip install -U -r requirements.txt -
Record the last working state:
pip freeze > requirements.deploy.txt
Warning: this project is nowhere near fully functional and should not under any circumstances be used in production. If you'd like to be contacted about beta testing opportunities, please PM /u/dakta.