FIX: Triage-flagged posts didn't have a score. (#752)

romanrizzi · web-flow · commit f789d3ee9686 · 2024-08-14T15:54:09.000-03:00
The score will contain the LLM result, and make sure the flag isn't displayed when a minimum score threshold is present.
diff --git a/config/locales/server.en.yml b/config/locales/server.en.yml
@@ -5,6 +5,10 @@ en:
         title: Triage posts using AI
         description: "Triage posts using a large language model"
         system_prompt_missing_post_placeholder: "System prompt must contain a placeholder for the post: %%POST%%"
+        flagged_post: |
+          <div>Response from the model:</div>
+          <p>%%LLM_RESPONSE%%</p>
+          <b>Triggered by the <a href="/admin/plugins/discourse-automation/%%AUTOMATION_ID%%">%%AUTOMATION_NAME%%</a> rule.</b>
       llm_report:
         title: Periodic report using AI
         description: "Periodic report based on a large language model"
diff --git a/discourse_automation/llm_triage.rb b/discourse_automation/llm_triage.rb
@@ -78,6 +78,7 @@
           canned_reply_user: canned_reply_user,
           hide_topic: hide_topic,
           flag_post: flag_post,
+          automation: self.automation,
         )
       rescue => e
         Discourse.warn_exception(e, message: "llm_triage: skipped triage on post #{post.id}")
diff --git a/lib/automation/llm_triage.rb b/lib/automation/llm_triage.rb
@@ -13,7 +13,8 @@ def self.handle(
         canned_reply: nil,
         canned_reply_user: nil,
         hide_topic: nil,
-        flag_post: nil
+        flag_post: nil,
+        automation: nil
       )
         if category_id.blank? && tags.blank? && canned_reply.blank? && hide_topic.blank? &&
              flag_post.blank?
@@ -38,12 +39,12 @@ def self.handle(
           llm.generate(
             filled_system_prompt,
             temperature: 0,
-            max_tokens: llm.tokenizer.tokenize(search_for_text).length * 2 + 10,
+            max_tokens: 700, # ~500 words
             user: Discourse.system_user,
             feature_name: "llm_triage",
-          )
+          )&.strip
 
-        if result.present? && result.strip.downcase.include?(search_for_text)
+        if result.present? && result.downcase.include?(search_for_text)
           user = User.find_by_username(canned_reply_user) if canned_reply_user.present?
           user = user || Discourse.system_user
           if canned_reply.present?
@@ -69,7 +70,24 @@ def self.handle(
 
           post.topic.update!(visible: false) if hide_topic
 
-          ReviewablePost.needs_review!(target: post, created_by: Discourse.system_user) if flag_post
+          if flag_post
+            reviewable =
+              ReviewablePost.needs_review!(target: post, created_by: Discourse.system_user)
+
+            score_reason =
+              I18n
+                .t("discourse_automation.scriptables.llm_triage.flagged_post")
+                .sub("%%LLM_RESPONSE%%", result)
+                .sub("%%AUTOMATION_ID%%", automation&.id.to_s)
+                .sub("%%AUTOMATION_NAME%%", automation&.name.to_s)
+
+            reviewable.add_score(
+              Discourse.system_user,
+              ReviewableScore.types[:needs_approval],
+              reason: score_reason,
+              force_review: true,
+            )
+          end
         end
       end
     end
diff --git a/spec/lib/modules/automation/llm_triage_spec.rb b/spec/lib/modules/automation/llm_triage_spec.rb
@@ -15,6 +15,7 @@ def triage(**args)
         hide_topic: true,
         system_prompt: "test %%POST%%",
         search_for_text: "bad",
+        automation: nil,
       )
     end
 
@@ -29,6 +30,7 @@ def triage(**args)
         hide_topic: true,
         system_prompt: "test %%POST%%",
         search_for_text: "bad",
+        automation: nil,
       )
     end
 
@@ -45,6 +47,7 @@ def triage(**args)
         category_id: category.id,
         system_prompt: "test %%POST%%",
         search_for_text: "bad",
+        automation: nil,
       )
     end
 
@@ -61,6 +64,7 @@ def triage(**args)
         search_for_text: "bad",
         canned_reply: "test canned reply 123",
         canned_reply_user: user.username,
+        automation: nil,
       )
     end
 
@@ -78,12 +82,14 @@ def triage(**args)
         system_prompt: "test %%POST%%",
         search_for_text: "bad",
         flag_post: true,
+        automation: nil,
       )
     end
 
     reviewable = ReviewablePost.last
 
     expect(reviewable.target).to eq(post)
+    expect(reviewable.reviewable_scores.first.reason).to include("bad")
   end
 
   it "can handle garbled output from LLM" do
@@ -94,6 +100,7 @@ def triage(**args)
         system_prompt: "test %%POST%%",
         search_for_text: "bad",
         flag_post: true,
+        automation: nil,
       )
     end
 

Original file line number	Diff line number	Diff line change
`@@ -78,6 +78,7 @@`
`78`	`78`	`canned_reply_user: canned_reply_user,`
`79`	`79`	`hide_topic: hide_topic,`
`80`	`80`	`flag_post: flag_post,`
	`81`	`+ automation: self.automation,`
`81`	`82`	`)`
`82`	`83`	`rescue => e`
`83`	`84`	`Discourse.warn_exception(e, message: "llm_triage: skipped triage on post #{post.id}")`