FIX: refine image and read command (#131)

- Attempt to hint reading is done by sending complete:true
- Do not include post_number in result unless it was sent in
- Rush visual feedback when a command is run (ensure we always revise)
- Include hyperlink in read command description
- Stop round tripping to GPT after image generation (speeds up images by a lot)
- Add a test for image command

config/locales/server.en.yml

@@ -102,7 +102,7 @@ en:
         google: "Search Google"
         read: "Read topic"
       command_description:
-        read: "Reading: %{title}"
+        read: "Reading: <a href='%{url}'>%{title}</a>"
         time: "Time in %{timezone} is %{time}"
         summarize: "Summarized <a href='%{url}'>%{title}</a>"
         image: "%{prompt}"

lib/modules/ai_bot/commands/command.rb

@@ -125,11 +125,13 @@ def invoke_and_attach_result_to(post, parent_post)
 
           raw = post.raw.sub(placeholder, raw)
 
+          post.revise(bot_user, { raw: raw }, skip_validations: true, skip_revision: true)
+
           if chain_next_response
+            # somewhat annoying but whitespace was stripped in revise
+            # so we need to save again
             post.raw = raw
             post.save!(validate: false)
-          else
-            post.revise(bot_user, { raw: raw }, skip_validations: true, skip_revision: true)
           end
 
           [chain_next_response, post]

lib/modules/ai_bot/commands/image_command.rb

@@ -15,20 +15,13 @@ def parameters
         [
           Parameter.new(
             name: "prompt",
-            description: "The prompt used to generate or create or draw the image",
+            description:
+              "The prompt used to generate or create or draw the image (40 words or less, be creative)",
             type: "string",
             required: true,
           ),
         ]
       end
-
-      def custom_system_message
-        <<~TEXT
-          In Discourse the markdown (description|SIZE, ZOOM%)[upload://SOMETEXT] is used to denote images and uploads. NEVER try changing the to http or https links.
-          ALWAYS prefer the upload:// format if available.
-          When rendering multiple images place them in a [grid] ... [/grid] block
-        TEXT
-      end
     end
 
     def result_name
@@ -40,7 +33,11 @@ def description_args
     end
 
     def chain_next_response
-      true
+      false
+    end
+
+    def custom_raw
+      @custom_raw
     end
 
     def process(prompt:)
@@ -58,7 +55,8 @@ def process(prompt:)
         f.unlink
       end
 
-      raw = <<~RAW
+      @custom_raw = <<~RAW
+
       [grid]
       #{
         uploads
@@ -68,7 +66,7 @@ def process(prompt:)
       [/grid]
     RAW
 
-      { prompt: prompt, markdown: raw, display_to_user: true }
+      { prompt: prompt, displayed_to_user: true }
     end
   end
 end

lib/modules/ai_bot/commands/read_command.rb

@@ -30,7 +30,7 @@ def parameters
     end
 
     def description_args
-      { title: @title }
+      { title: @title, url: @url }
     end
 
     def process(topic_id:, post_number: nil)
@@ -46,6 +46,8 @@ def process(topic_id:, post_number: nil)
       @title = topic.title
 
       posts = Post.secured(Guardian.new).where(topic_id: topic_id).order(:post_number).limit(40)
+      @url = topic.relative_url(post_number)
+
       posts = posts.where("post_number = ?", post_number) if post_number
 
       content = +"title: #{topic.title}\n\n"
@@ -55,7 +57,9 @@ def process(topic_id:, post_number: nil)
       # TODO: 16k or 100k models can handle a lot more tokens
       content = ::DiscourseAi::Tokenizer::BertTokenizer.truncate(content, 1500).squish
 
-      { topic_id: topic_id, post_number: post_number, content: content }
+      result = { topic_id: topic_id, content: content, complete: true }
+      result[:post_number] = post_number if post_number
+      result
     end
   end
 end

spec/lib/modules/ai_bot/commands/google_command_spec.rb

@@ -1,7 +1,5 @@
 #frozen_string_literal: true
 
-require_relative "../../../../support/openai_completions_inference_stubs"
-
 RSpec.describe DiscourseAi::AiBot::Commands::GoogleCommand do
   fab!(:bot_user) { User.find(DiscourseAi::AiBot::EntryPoint::GPT3_5_TURBO_ID) }
 

spec/lib/modules/ai_bot/commands/image_command_spec.rb

@@ -0,0 +1,36 @@
+#frozen_string_literal: true
+
+RSpec.describe DiscourseAi::AiBot::Commands::ImageCommand do
+  fab!(:bot_user) { User.find(DiscourseAi::AiBot::EntryPoint::GPT3_5_TURBO_ID) }
+
+  describe "#process" do
+    it "can generate correct info" do
+      post = Fabricate(:post)
+
+      SiteSetting.ai_stability_api_url = "https://api.stability.dev"
+      SiteSetting.ai_stability_api_key = "abc"
+
+      image =
+        "iVBORw0KGgoAAAANSUhEUgAAAAEAAAABCAYAAAAfFcSJAAAADUlEQVR42mP8z8BQDwAEhQGAhKmMIQAAAABJRU5ErkJggg=="
+
+      stub_request(
+        :post,
+        "https://api.stability.dev/v1/generation/#{SiteSetting.ai_stability_engine}/text-to-image",
+      )
+        .with do |request|
+          json = JSON.parse(request.body)
+          expect(json["text_prompts"][0]["text"]).to eq("a pink cow")
+          true
+        end
+        .to_return(status: 200, body: { artifacts: [{ base64: image }, { base64: image }] }.to_json)
+
+      image = described_class.new(bot_user, post)
+      info = image.process(prompt: "a pink cow").to_json
+
+      expect(JSON.parse(info)).to eq("prompt" => "a pink cow", "displayed_to_user" => true)
+      expect(image.custom_raw).to include("upload://")
+      expect(image.custom_raw).to include("[grid]")
+      expect(image.custom_raw).to include("a pink cow")
+    end
+  end
+end

spec/lib/modules/ai_bot/commands/read_command_spec.rb

@@ -15,6 +15,7 @@
       expect(results[:topic_id]).to eq(post1.topic_id)
       expect(results[:content]).to include("hello")
       expect(results[:content]).to include("sam")
+      expect(read.description_args).to eq(title: post1.topic.title, url: post1.topic.relative_url)
     end
   end
 end