backend for test button

Author: SamSaffron

app/controllers/discourse_ai/admin/ai_spam_controller.rb

@@ -47,6 +47,35 @@ def update
         render json: AiSpamSerializer.new(spam_config, root: false)
       end
 
+      def test
+        url = params[:post_url].to_s
+        post = nil
+
+        if url.match?(/^\d+$/)
+          post_id = url.to_i
+          post = Post.find_by(id: post_id)
+        end
+
+        route = UrlHelper.rails_route_from_url(url) if !post
+
+        if route
+          if route[:controller] == "topics"
+            post_number = route[:post_number] || 1
+            post = Post.with_deleted.find_by(post_number: post_number, topic_id: route[:topic_id])
+          end
+        end
+
+        raise Discourse::NotFound if !post
+
+        result =
+          DiscourseAi::AiModeration::SpamScanner.test_post(
+            post,
+            custom_instructions: params[:custom_instructions],
+          )
+
+        render json: result
+      end
+
       private
 
       def allowed_params

config/routes.rb

@@ -81,6 +81,7 @@
     get "/ai-usage", to: "discourse_ai/admin/ai_usage#show"
     get "/ai-spam", to: "discourse_ai/admin/ai_spam#show"
     put "/ai-spam", to: "discourse_ai/admin/ai_spam#update"
+    post "/ai-spam/test", to: "discourse_ai/admin/ai_spam#test"
 
     resources :ai_llms,
               only: %i[index create show update destroy],

lib/ai_moderation/spam_scanner.rb

@@ -119,26 +119,68 @@ def self.significant_change?(previous_version, current_version)
         distance >= MINIMUM_EDIT_DIFFERENCE
       end
 
-      def self.perform_scan(post)
-        return if !enabled?
-        return if !should_scan_post?(post)
-
+      def self.test_post(post, custom_instructions: nil)
         settings = AiModerationSetting.spam
-        return if !settings || !settings.llm_model
-
         llm = settings.llm_model.to_llm
-        custom_instructions = settings.custom_instructions.presence
+        custom_instructions = custom_instructions || settings.custom_instructions.presence
+        context = build_context(post)
+        prompt = completion_prompt(post, context: context, custom_instructions: custom_instructions)
 
-        system_prompt = build_system_prompt(custom_instructions)
-        prompt = DiscourseAi::Completions::Prompt.new(system_prompt)
+        result =
+          llm.generate(
+            prompt,
+            temperature: 0.1,
+            max_tokens: 100,
+            user: Discourse.system_user,
+            feature_name: "spam_detection_test",
+            feature_context: {
+              post_id: post.id,
+            },
+          )&.strip
+
+        history = nil
+        AiSpamLog.where(post: post).order(:created_at).limit(100).each do |log|
+          history ||= +"Scan History:\n"
+          history << "date: #{log.created_at} is_spam: #{log.is_spam}\n"
+        end
 
-        context = build_context(post)
+        log = +"Scanning #{post.url}\n\n"
+
+        if history
+          log << history
+          log << "\n"
+        end
+
+        log << "LLM: #{settings.llm_model.name}\n\n"
+        log << "System Prompt: #{build_system_prompt(custom_instructions)}\n\n"
+        log << "Context: #{context}\n\n"
+        log << "Result: #{result}"
+
+        is_spam = (result.present? && result.downcase.include?("spam"))
+        { is_spam: is_spam, log: log }
+      end
 
+      def self.completion_prompt(post, context:, custom_instructions:)
+        system_prompt = build_system_prompt(custom_instructions)
+        prompt = DiscourseAi::Completions::Prompt.new(system_prompt)
         args = { type: :user, content: context }
         upload_ids = post.upload_ids
         args[:upload_ids] = upload_ids.take(3) if upload_ids.present?
-
         prompt.push(**args)
+        prompt
+      end
+
+      def self.perform_scan(post)
+        return if !enabled?
+        return if !should_scan_post?(post)
+
+        settings = AiModerationSetting.spam
+        return if !settings || !settings.llm_model
+
+        context = build_context(post)
+        llm = settings.llm_model.to_llm
+        custom_instructions = settings.custom_instructions.presence
+        prompt = completion_prompt(post, context: context, custom_instructions: custom_instructions)
 
         begin
           result =
@@ -156,7 +198,6 @@ def self.perform_scan(post)
           is_spam = (result.present? && result.downcase.include?("spam"))
 
           log = AiApiAuditLog.order(id: :desc).where(feature_name: "spam_detection").first
-
           AiSpamLog.transaction do
             log =
               AiSpamLog.create!(

spec/requests/admin/ai_spam_controller_spec.rb

@@ -79,6 +79,78 @@
     end
   end
 
+  describe "#test" do
+    fab!(:spam_post) { Fabricate(:post) }
+    fab!(:spam_post2) { Fabricate(:post, topic: spam_post.topic, raw: "something special 123") }
+    fab!(:setting) do
+      AiModerationSetting.create(
+        {
+          setting_type: :spam,
+          llm_model_id: llm_model.id,
+          data: {
+            custom_instructions: "custom instructions",
+          },
+        },
+      )
+    end
+
+    before { sign_in(admin) }
+
+    it "can scan using post url" do
+      DiscourseAi::Completions::Llm.with_prepared_responses(["spam"]) do
+        post "/admin/plugins/discourse-ai/ai-spam/test.json", params: { post_url: spam_post2.url }
+      end
+
+      expect(response.status).to eq(200)
+
+      parsed = response.parsed_body
+      expect(parsed["log"]).to include(spam_post2.raw)
+    end
+
+    it "can scan using post id" do
+      DiscourseAi::Completions::Llm.with_prepared_responses(["spam"]) do
+        post "/admin/plugins/discourse-ai/ai-spam/test.json",
+             params: {
+               post_url: spam_post.id.to_s,
+             }
+      end
+
+      expect(response.status).to eq(200)
+
+      parsed = response.parsed_body
+      expect(parsed["log"]).to include(spam_post.raw)
+    end
+
+    it "returns proper spam test results" do
+      freeze_time DateTime.parse("2000-01-01")
+
+      AiSpamLog.create!(
+        post: spam_post,
+        llm_model: llm_model,
+        is_spam: false,
+        created_at: 2.days.ago,
+      )
+
+      AiSpamLog.create!(post: spam_post, llm_model: llm_model, is_spam: true, created_at: 1.day.ago)
+
+      DiscourseAi::Completions::Llm.with_prepared_responses(["spam"]) do
+        post "/admin/plugins/discourse-ai/ai-spam/test.json",
+             params: {
+               post_url: spam_post.url,
+               custom_instructions: "special custom instructions",
+             }
+      end
+
+      expect(response.status).to eq(200)
+
+      parsed = response.parsed_body
+      expect(parsed["log"]).to include("special custom instructions")
+      expect(parsed["log"]).to include(spam_post.raw)
+      expect(parsed["is_spam"]).to eq(true)
+      expect(parsed["log"]).to include("Scan History:")
+    end
+  end
+
   describe "#show" do
     context "when logged in as admin" do
       before { sign_in(admin) }
@@ -132,19 +204,6 @@
 
         expect(json["flagging_username"]).to eq(flagging_user.username)
       end
-
-      it "includes the stats" do
-        expect(SpamScanner.flagging_user.id).not_to eq(SystemUser.id)
-        get "/admin/plugins/discourse-ai/ai-spam.json"
-
-        json = response.parsed_body
-        expect(json["stats"]).to include(
-          "scanned_count",
-          "spam_detected",
-          "false_positives",
-          "false_negatives",
-        )
-      end
     end
 
     context "when not logged in as admin" do