I want to make a PR to discourse/discourse-ai that changes the behavior of ai bot.
Can you help me create this PR, I am going to need tests as well.
Start by searching through the repo so you find the names of all the files involved, then read the important files.
Searched for 'ai bot PM reply' in discourse/discourse-ai
Searched for 'reply as other persona' in discourse/discourse-ai
Retrieved content of lib/ai_bot/playground.rb, lib/ai_bot/entry_point.rb, spec/lib/modules/ai_bot/playground_spec.rb, spec/lib/modules/ai_bot/entry_point_spec.rb from discourse/discourse-ai@main
Based on the requirement to change the behavior of the AI bot in the discourse/discourse-ai repository as per your specifications, I found the relevant files involved in handling replies in the AI bot Private Messages (PMs). Below are the summaries and recommendations for modification:
playground.rb (lib/ai_bot/playground.rb)
schedule_reply determines when and how AI bot or other personas should reply to PMs based on the conditions outlined in your request.schedule_reply to check if the reply is directly to another user or to another persona. This can be determined by analyzing the mentions and topic_allowed_users to distinguish replies to users from replies to personas.schedule_reply to skip AI bot replies when a user replies directly to another user in an AI bot PM.entry_point.rb (lib/ai_bot/entry_point.rb)
playground.rb are compatible with the bot initialization and event handling described in this file.playground_spec.rb (spec/lib/modules/ai_bot/playground_spec.rb)
entry_point_spec.rb (spec/lib/modules/ai_bot/entry_point_spec.rb)
schedule_reply method in the playground.rb file. Ensure that your logic correctly identifies the context of the reply (direct user reply vs. reply to persona) and alters the behavior of the AI bot accordingly.playground_spec.rb. Create test cases that mimic the scenarios outlined in your requirements (direct replies to users vs. replies to other personas) and validate the expected outcomes.Before starting the implementation, it’s advisable to map out the precise logic and conditions for identifying replies to users vs. personas. This will ensure that the changes are accurately targeted and effective. Additionally, continuously run the existing and new tests to ensure that modifications do not introduce regressions or unintended behavior changes in the AI bot functionality.
OK start with the changes to schedule_reply please.
To address the requirements for modifying the AI bot’s behavior in response to user replies within private messages (PMs), here is a proposed approach to adapt the schedule_reply method in the lib/ai_bot/playground.rb file. This method will be adjusted to incorporate the logic needed to differentiate between replies directly to another user and replies to another persona within an AI bot PM. Based on this differentiation, the AI bot will either abstain from responding (in the case of direct user replies) or it will reply as another persona (when the reply is directed at a persona).
Below is an example of how you might adapt the schedule_reply method to accommodate these requirements. Please note that this is a high-level outline intended to guide the implementation. Actual code might require adjustments based on the complete context of the application and additional logic in the playground.rb file.
module DiscourseAi
module AiBot
class Playground
def self.schedule_reply(post)
return if is_bot_user_id?(post.user_id)
bot_ids = DiscourseAi::AiBot::EntryPoint::BOT_USER_IDS
mentionables = AiPersona.mentionables(user: post.user)
# Identify if the reply is to another user directly or to a persona
direct_reply_to_user = post.reply_to_user.present?
mentioned_persona = nil
if post.topic.private_message?
mentioned_persona = if mentionables.present?
# extracting mentions from the post
mentions = post.raw_mentions
# Find if a mentioned user is a persona
mentionables.find { |mentionable| mentions.include?(mentionable[:username].downcase) }
end
end
# Conditions to modify behavior based on the target of the reply
if direct_reply_to_user
# If this is a direct user reply in an AI bot PM, do not schedule a reply from the bot.
return
elsif mentioned_persona
# This section is entered if a reply is to another persona; here we need to schedule a reply as another persona
bot_user = User.find(mentioned_persona[:user_id])
bot = DiscourseAi::AiBot::Bot.as(bot_user, persona: mentioned_persona.class_instance.new)
new(bot).update_playground_with(post)
else
# Default behavior (existing behavior) if not replying directly to a user or mentioning a persona
default_bot_reply(post, bot_ids, mentionables)
end
end
private
# This method encapsulates the existing behavior to allow for clear separation of the new conditional logic
def self.default_bot_reply(post, bot_ids, mentionables)
# Existing logic to handle replies in AI bot PMs
end
# Existing methods not shown for brevity
end
end
end
Steps for Implementation:
schedule_reply method, ensuring it works cohesively with any existing conditions and actions.playground_spec.rb to simulate scenarios where users reply directly to another user or to a persona within an AI bot PM.This approach allows for a clear distinction between different types of replies in PMs and adjusts the AI bot’s behavior to either respond or not, based on the identified reply context.
but if we are replying to a bot it is fine we can continue …
Given the clarification that replying to a bot (including AI bot personas) should continue the conversation seamlessly, we need to adjust the previously proposed changes to the schedule_reply method. This adjustment ensures that:
Here’s a refined approach to adapt the schedule_reply method in the lib/ai_bot/playground.rb file:
module DiscourseAi
module AiBot
class Playground
def self.schedule_reply(post)
return if is_bot_user_id?(post.user_id)
bot_ids = DiscourseAi::AiBot::EntryPoint::BOT_USER_IDS
mentionables = AiPersona.mentionables(user: post.user)
if post.topic.private_message?
mentioned_users = post.raw_mentions.map(&:downcase)
# Identifies if the reply is directly to a non-bot user by checking mentionables and bot_ids
direct_reply_to_non_bot_user = mentioned_users.present? &&
!(mentioned_users & mentionables.map { |m| m[:username].downcase }).empty? &&
!(mentioned_users & User.where(id: bot_ids).pluck(:username).map(&:downcase)).empty?
if direct_reply_to_non_bot_user
# If this is a direct reply to a non-bot user in an AI bot PM, do not schedule a bot reply.
return
end
# For all other cases, including no direct mentions or mentions of bots/personas, proceed with regular bot scheduling
bot_user = determine_bot_user(post, bot_ids, mentionables)
bot = DiscourseAi::AiBot::Bot.as(bot_user)
new(bot).update_playground_with(post)
end
end
private
def self.determine_bot_user(post, bot_ids, mentionables)
# Logic to determine the appropriate bot user based on mentions or default to the AI bot.
# This method returns the bot User object.
# Example implementation, adjust based on actual bot selection logic:
mentioned_bot_id = (post.raw_mentions & User.where(id: bot_ids).pluck(:username)).first
return User.find_by(username: mentioned_bot_id) if mentioned_bot_id
# If no specific bot is mentioned, select based on other criteria or defaults
default_bot_user_id = bot_ids.include?(post.topic.custom_fields["ai_persona_id"]) ? post.topic.custom_fields["ai_persona_id"].to_i : bot_ids.first
User.find_by(id: default_bot_user_id)
end
# Existing methods not shown for brevity
end
end
end
Key Adjustments:
direct_reply_to_non_bot_user is introduced to determine if the reply is specifically directed (mentioned) to non-bot users. This effectively bypasses the bot response if the condition is true.determine_bot_user method serves to explicitly choose the bot user that should reply. This method can be tailored further to incorporate complex logic for selecting among multiple bots/personas based on mentions or other criteria.Next Steps:
schedule_reply method with the adjusted logic provided above, ensuring it fits seamlessly with existing code. Update or add tests in playground_spec.rb to cover the new logic paths—particularly ensuring that replies to non-bot users in PMs don’t trigger bot responses, while replies to bots/personas do.This approach ensures a nuanced handling of user interactions within PMs, promoting a natural flow of conversation with bot personas while preventing unnecessary bot responses to direct, non-bot user interactions.
OK assume I made the changes I needed, make an integration test that confirms the new behavior
Assuming you’ve implemented the necessary changes to the schedule_reply method in the lib/ai_bot/playground.rb file, the next step is to create an integration test that validates the new behavior of the AI bot. This test will confirm that:
Let’s outline an integration test scenario using RSpec, located in spec/lib/modules/ai_bot/playground_spec.rb, which confirms these behaviors:
require 'rails_helper'
RSpec.describe DiscourseAi::AiBot::Playground do
fab!(:bot_user) { User.find(-110) } # Assuming -110 is a bot's user ID for the sake of the example
fab!(:persona_user) { Fabricate(:user) } # Assuming this user is configured as a persona
fab!(:regular_user) { Fabricate(:user) }
let(:ai_bot_playground) { DiscourseAi::AiBot::Playground }
before do
SiteSetting.ai_bot_enabled = true
# Additional setup as necessary for enabling AI bot, personas, etc.
end
describe '#schedule_reply behavior' do
let!(:pm) do
Fabricate(
:private_message_topic,
user: regular_user,
topic_allowed_users: [
Fabricate.build(:topic_allowed_user, user: regular_user),
Fabricate.build(:topic_allowed_user, user: bot_user),
Fabricate.build(:topic_allowed_user, user: persona_user)
]
)
end
it 'does not reply to a direct reply to a regular user' do
original_post_count = pm.posts.count
# Simulate a post by a regular user directly replying to another regular user
PostCreator.create!(
regular_user,
raw: "Thanks for your input, @#{regular_user.username}",
topic_id: pm.id
)
# Assuming background jobs run synchronously for the sake of this example
expect(pm.reload.posts.count).to eq(original_post_count + 1) # Only the user's new post, no bot reply
end
it 'replies to a direct mention of a persona bot in PM' do
original_post_count = pm.posts.count
# Simulate a post mentioning the persona bot
PostCreator.create!(
regular_user,
raw: "Hey @#{persona_user.username}, can you give me more information?",
topic_id: pm.id
)
# Assuming background jobs run synchronously for the sake of this example
expect(pm.reload.posts.count).to eq(original_post_count + 2) # Includes the new post and bot's (persona's) reply
end
end
end
Key Points in This Example:
fab!(:bot_user) and fab!(:persona_user) lines presume specific user setups. You’ll need to adjust these to match your actual bot and persona user configurations.sidekiq/testing inline mode for Sidekiq).Next Steps: