sobotics/firealarm

A Stack Exchange chatbot to catch low-quality posts.

firealarm's Issues

The bot just leaked its API key

This message in FireAlarm Development contains the API key of FireAlarm. Is this bad enough to warrant changing the key?

Restrict the train wreck for just certain situations

As I mentioned here #19 (comment):

I'd rather say that we remove the train wreck for the normal jobs which we do daily, like feedbacks. The #rekdtrain shouldn't fire when we're giving a bunch of feedback to any of the bots, either using the feedback command, or the reply option. Let the train wreck remain when a lot of people say the same general thing.

The issue which @danbopes is facing is that, they're seeing the #rekdtrain almost daily, at which point the joke becomes annoying.

Add a tag blacklist

Some tags attract almost nothing but bad questions, for example [appstore-approval]. I'm going to add a tag blacklist, so tags like these can be monitored effectively.

FireAlarm reports closed questions.

The title explains it. Posting it here as a reminder for myself.

Add the train wrecker again

The train wrecker was recently removed, but on SOBotics, it appears that most users want it back.

Can we add it back?

Punctuation suffix to commands should be allowed

Commands like alive? should automatically run alive.

Reduce the rate of questions feed into the room by introducing a delay and checking for close votes

The rate of question the bot finds is a little high. I am wondering if we could mitigate this by putting a delay between when the question is posted and when the bot notifies the room. If we could have it wait X minutes (I'm thinking like 15-30 minutes) and then if there are no close votes then go ahead and notify. If it has close votes or is already closed then the tag is working on it or has done its job and there is no reason to let us know.

Just like [cv-pls] we should be giving the people in the tag the opportunity to handle the trash before we step in.

Links to questions containing "[" are broken

Example: http://chat.stackoverflow.com/transcript/message/35731300#35731300

Solution: Escaping [ and ] will work. (I would not change this in SwiftStack, since this is just an issue when using markdown)

Enhancements

I plan to do some enhancements in FA-Swift.

• Only autoupdate when the commit message contains --autoupdate (something like Smokey), otherwise update when a user manually runs the update command.

• A command called leave room which will just leave the room instead of being killed in every other room too.

• Print the likelihood of a report for debugging purposes.

• Make the start message contain more information.

• Location command

• Print the likelihood when a check post fails.

• report command to report a post (mainly for testing purposes). Requires the owner privilege.

• Rewrite wikis

• A test command to allow testing of given text.

• Add a command to show which all room FireAlarm is in.

• Add a why command.

• Add a membership command printing all privileged users.

• Add a enhancement so that we can check the privs of any other user from chat.

• Add a CI

Remove trailing backslashes from question titles

Today we had a report where the question's title ended with a backslash (\), this caused the link markdown to be escaped and thus broke the link:

[ FireAlarm-Swift ] [tag:php] Potentially bad question [Undefined index: files in C:](//stackoverflow.com/q/49947530)

The question's title:

Undefined index: files in C:\

Sadly it seems you can't escape backslashes in chat, so they will have to be removed.

Ideas for new filters

FireAlarm currently uses a Naive Bayes classifier to detect bad posts. This works great, but a couple of other filters which add to the weight would help detect a lot more bad posts. If you have an idea for a new filter, please post one! One per comment, please.

The Naive Bayes filter, Misleading Links, and Blacklisted tag/keyword/user filters already exist.

Do not reply to feedback on FireAlarm report

It feels natural to reply on FireAlarm report (even if it's not official feedback), mostly to notify other people in room.

I suggest to remove the response as in example

Unrecognized command mcve; did you mean alive?

See https://chat.stackoverflow.com/transcript/message/43711697#43711697 for more context.

Whenever an error occurs, the bot pings the hardcoded user instead of the owner of the instance

For example, right now it is hardcoded to ping NobodyNada right now in case of an errors. Instead of this, it should ping the instance owner on an error.