completion-kit 0.5.12 → 0.5.14

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 9c45ef9dab51631cbc2bfc606b6c98c038d967edeea8903efa634571d012e0b0
-  data.tar.gz: 549981db86a07f2d48cafb2b2413f6e1850997087ef553418c3ec9d7d07e0a0a
+  metadata.gz: 119b2978f4176a40950e2874176272851530c0b00a1a5dc9387f76ffd8caba62
+  data.tar.gz: ca60f782bcd182e9af024d34479e78f21c85b8e8605ade8f825d00cdc448c4ea
 SHA512:
-  metadata.gz: ddf258f7c59bc3a92113bcb95211f0314ea6cebe624d163478833f1515f0baafab0c630b9dd8983b7e2bde7ea5b641cb45b2249cd48c08a853422633e93409d9
-  data.tar.gz: 5001137b6be3a376f665cff46480e51b6994774f7e9146b5e464053bedcd5d989e2171b4ea8d792c9312b8c03b2e46e9d6b7515d8037e1c5510bde14caf11b1f
+  metadata.gz: 1ce0a05b7b22f363f0386fffa2b94c60b94e767ac0a732b1309d70dbda9d951407f5abc42d02e3992f56a20bc902f9bcba938dedd405cfd02cb8577a0b8d4fe5
+  data.tar.gz: 0ed8728ca2853047cd6c659d696cdfbbeac681f726d3d368e183187ac179f434933a6f0d3b819e4bb5318f5d643aebbd4991a92014e7117993cee9200a6cca37

data/app/assets/stylesheets/completion_kit/application.css

@@ -39,6 +39,7 @@
 }
 
 html {
+  color-scheme: dark;
   font-size: 18px;
   line-height: 1.5;
   scrollbar-gutter: stable;
@@ -1284,6 +1285,24 @@ tr:hover .ck-chip--publish {
   line-height: 1.75;
 }
 
+/* Scrollable code-block wrapper — caps height for big payloads. Inner
+   <pre> drops its own border/radius so the wrapper owns the edge. We
+   leave the scrollbar unstyled so it matches the browser/page scrollbar. */
+.ck-code-scroll-wrap {
+  margin-top: 0.5rem;
+  max-height: 28rem;
+  overflow: auto;
+  border: 1px solid var(--ck-line);
+  border-radius: var(--ck-radius);
+  background: var(--ck-bg-strong);
+}
+.ck-code-scroll-wrap > .ck-code {
+  margin-top: 0;
+  border: 0;
+  border-radius: 0;
+  background: transparent;
+}
+
 .ck-note-box {
   background: var(--ck-surface-soft);
   border: 1px solid var(--ck-line);
@@ -1873,22 +1892,7 @@ tr:hover .ck-chip--publish {
   background: var(--ck-bg-strong);
   overflow: auto;
   max-height: 60vh;
-  scrollbar-width: thin;
-  scrollbar-color: var(--ck-line-strong) var(--ck-bg);
 }
-.ck-csv-table-wrap::-webkit-scrollbar { width: 12px; height: 12px; background: var(--ck-bg); }
-.ck-csv-table-wrap::-webkit-scrollbar-track {
-  background: var(--ck-bg);
-  border-left: 1px solid var(--ck-line);
-  border-top: 1px solid var(--ck-line);
-}
-.ck-csv-table-wrap::-webkit-scrollbar-thumb {
-  background: var(--ck-line-strong);
-  border-radius: 6px;
-  border: 3px solid var(--ck-bg);
-}
-.ck-csv-table-wrap::-webkit-scrollbar-thumb:hover { background: var(--ck-muted); }
-.ck-csv-table-wrap::-webkit-scrollbar-corner { background: var(--ck-bg); }
 
 .ck-modal__body .ck-csv-table-wrap {
   margin-top: 0;
@@ -3197,6 +3201,7 @@ a.ck-metric-group-pill {
   white-space: nowrap;
 }
 
+
 .ck-runs-table th {
   vertical-align: middle;
   white-space: nowrap;
@@ -3208,21 +3213,44 @@ a.ck-metric-group-pill {
   padding-bottom: 0.7rem;
 }
 
+.ck-runs-table th:first-child,
+.ck-runs-table td:first-child {
+  width: 1%;
+}
+
 .ck-runs-table th:nth-child(n+2),
 .ck-runs-table td:nth-child(n+2) {
   width: 1%;
   white-space: nowrap;
 }
 
+.ck-runs-table th:last-child,
+.ck-runs-table td:last-child {
+  width: auto;
+}
+
 .ck-runs-table__identity {
   display: flex;
   flex-direction: column;
   gap: 0.25rem;
   min-width: 0;
+  width: 27rem;
+  max-width: 100%;
 }
 
 .ck-runs-table .ck-run-name {
   line-height: 1.2;
+  display: flex;
+  align-items: center;
+  gap: 0.5rem;
+  min-width: 0;
+}
+
+.ck-runs-table .ck-run-name strong {
+  overflow: hidden;
+  text-overflow: ellipsis;
+  white-space: nowrap;
+  min-width: 0;
 }
 
 .ck-runs-table__config {

data/app/helpers/completion_kit/application_helper.rb

@@ -179,6 +179,18 @@ module CompletionKit
       ["tag", "tag-#{tag.color}", ("tag-outline" if outline)].compact.join(" ")
     end
 
+    def ck_format_maybe_json(text)
+      s = text.to_s
+      return s if s.strip.empty?
+      first = s.strip[0]
+      return s unless first == "{" || first == "["
+      begin
+        JSON.pretty_generate(JSON.parse(s))
+      rescue JSON::ParserError
+        s
+      end
+    end
+
     def tag_filter_url(base_path, selected, toggling)
       remaining = selected.reject { |t| t.id == toggling.id }
       next_set = selected.include?(toggling) ? remaining : remaining + [toggling]

data/app/services/completion_kit/open_ai_client.rb

@@ -15,7 +15,7 @@ module CompletionKit
       return "Error: API key not configured" unless configured?
 
       model = options[:model] || "gpt-4.1-mini"
-      max_tokens = options[:max_tokens] || 1000
+      max_tokens = options[:max_tokens] || 8192
       temperature = options[:temperature] || 0.7
 
       response = post_responses(model: model, prompt: prompt, max_tokens: max_tokens, temperature: temperature)
@@ -36,8 +36,14 @@ module CompletionKit
 
       if response.success?
         data = JSON.parse(response.body)
+        if data["status"] == "incomplete"
+          reason = data.dig("incomplete_details", "reason") || "unknown"
+          return "Error: response incomplete (#{reason}) — increase max_tokens=#{max_tokens} or pick a non-reasoning judge model"
+        end
         message = Array(data["output"]).find { |o| o["type"] == "message" }
-        message&.dig("content", 0, "text").to_s.strip
+        content = message&.dig("content", 0, "text").to_s.strip
+        return "Error: model returned empty content" if content.empty?
+        content
       else
         "Error: #{response.status} - #{response.body}"
       end

data/app/services/completion_kit/open_router_client.rb

@@ -13,7 +13,7 @@ module CompletionKit
       return "Error: API key not configured" unless configured?
 
       model = options[:model] || "openai/gpt-4o-mini"
-      max_tokens = options[:max_tokens] || 1000
+      max_tokens = options[:max_tokens] || 8192
       temperature = options[:temperature] || 0.7
 
       response = post_chat(model: model, prompt: prompt, max_tokens: max_tokens, temperature: temperature)
@@ -34,7 +34,13 @@ module CompletionKit
 
       if response.success?
         data = JSON.parse(response.body)
-        data.dig("choices", 0, "message", "content").to_s.strip
+        choice = data.dig("choices", 0) || {}
+        if choice["finish_reason"] == "length"
+          return "Error: response truncated by max_tokens=#{max_tokens} before visible content was emitted (reasoning model burned through the budget)"
+        end
+        content = choice.dig("message", "content").to_s.strip
+        return "Error: model returned empty content" if content.empty?
+        content
       else
         "Error: #{response.status} - #{response.body}"
       end

data/app/views/completion_kit/responses/show.html.erb

@@ -67,7 +67,7 @@
 
 <section>
   <p class="ck-kicker">Input</p>
-  <pre class="ck-code ck-code--dark"><%= begin; JSON.pretty_generate(JSON.parse(@response.input_data)); rescue; @response.input_data; end %></pre>
+  <div class="ck-code-scroll-wrap"><pre class="ck-code"><%= ck_format_maybe_json(@response.input_data) %></pre></div>
 </section>
 
 <section class="ck-card--spaced">
@@ -77,13 +77,13 @@
       <span class="ck-chip ck-chip--soft" style="text-transform: none;"><%= @run.prompt.llm_model %></span>
     <% end %>
   </div>
-  <pre class="ck-code"><%= @response.response_text %></pre>
+  <div class="ck-code-scroll-wrap"><pre class="ck-code"><%= ck_format_maybe_json(@response.response_text) %></pre></div>
 </section>
 
 <% if @response.expected_output.present? %>
   <section class="ck-card--spaced">
     <p class="ck-kicker">Expected</p>
-    <pre class="ck-code"><%= @response.expected_output %></pre>
+    <div class="ck-code-scroll-wrap"><pre class="ck-code"><%= ck_format_maybe_json(@response.expected_output) %></pre></div>
   </section>
 <% end %>
 

data/lib/completion_kit/version.rb

@@ -1,3 +1,3 @@
 module CompletionKit
-  VERSION = "0.5.12"
+  VERSION = "0.5.14"
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: completion-kit
 version: !ruby/object:Gem::Version
-  version: 0.5.12
+  version: 0.5.14
 platform: ruby
 authors:
 - Damien Bastin