actionagent 1.9.3 → 1.9.5

data/app/services/action_agent/evaluation_report_import.rb

@@ -17,9 +17,18 @@ module ActionAgent
   #   evaluation — that agent's evaluation for the `suite` and the report's
   #                scope (SCOPE_KEYS in its metadata), named for both
   #   scenarios  — one per reported scenario key, updated to the reported prompt
-  #   run        — one complete EvaluationRun per tenant and run_id, with one
+  #   run        — one EvaluationRun per tenant and run_id, with one
   #                EvaluationScenarioResult per scenario and model
   #
+  # An envelope's `status` is "complete" when absent. A "running" envelope
+  # stores the results so far as a running run, and each later delivery under
+  # its run_id replaces them, until a "complete" one finishes the run:
+  #   running  — "running" or "complete" replaces the stored results
+  #   complete — "complete" is an identical retry or a Conflict, as for any
+  #              stored run; "running" changes nothing, so a progress
+  #              delivery that lands after the final one is harmless
+  # Only the first delivery under a run_id is admitted and counted.
+  #
   # The run's per-model summary, criterion scores and recommendations are
   # computed from the stored results, not taken from the report. The judge's
   # verdict and label are taken from it, and so is what the judge spent —
@@ -85,6 +94,11 @@ module ActionAgent
     MAX_JUDGE_KINDS = 10
     SCOPE_PATTERN = %r{\A[\w .:/@-]{1,100}\z}
     STATUSES = %w[passed failed errored].freeze
+    # The envelope statuses a delivery may name.
+    RUN_STATUSES = %w[running complete].freeze
+    # What this collector accepts beyond a version-1 report, sent with its
+    # refusal of a payload that is not one so a publisher can ask before a run.
+    CAPABILITIES = %w[progress].freeze
     # Report metadata that tells one evaluation of a suite from another, in the
     # order it is written into the evaluation's name.
     SCOPE_KEYS = %w[scope environment role].freeze
@@ -102,20 +116,22 @@ module ActionAgent
     # The evaluation_runs columns an import writes that later migrations added.
     REQUIRED_RUN_COLUMNS = %w[external_tenant external_run_id external_report_digest agent_version_id].freeze
 
-    # Returns `[run, duplicate]`: the stored EvaluationRun, and whether an
-    # earlier identical delivery had already stored it. `account` is the
-    # tenant, or nil on a single-tenant install.
+    # Returns `[run, duplicate, progressed]`: the stored EvaluationRun, whether
+    # an earlier delivery had already stored this report, and whether this
+    # delivery replaced the results of a running run. `account` is the tenant,
+    # or nil on a single-tenant install.
     #
-    # `admit` is asked only for a report that would be stored, never for an
-    # identical retry. It returns nil to allow the report, or a denial, which
-    # is raised as Denied.
+    # `admit` is asked only for a run_id the tenant holds no run under. It
+    # returns nil to allow the report, or a denial, which is raised as Denied.
     #
     # The envelope's identities are validated as sent. NUL characters, which
     # PostgreSQL cannot store, are removed from every string inside `report`.
     #
     # Raises Invalid for a payload that is not a valid version-1 report, or an
     # evaluation name the agent already holds for another suite or scope;
-    # Conflict for different content under a stored run_id; Refused (and
+    # Conflict for different content under a complete run_id, or a delivery
+    # under a running run_id that the dashboard executed or that names another
+    # agent, suite or scope; Refused (and
     # LimitExceeded) when storing needs an operator to act first; Denied; and
     # Busy.
     def self.call(payload:, account: nil, admit: nil)
@@ -162,24 +178,25 @@ module ActionAgent
       validate!
       digest = self.class.digest(@payload)
       existing = stored_run
-      return [ identical!(existing, digest), true ] if existing
+      return settled(existing, digest) if settles?(existing, digest)
 
       attempts = 0
       begin
         OwnerLock.new(EvaluationRun.connection, owner_lock_key).synchronize do
           EvaluationRun.transaction(requires_new: true) do
             existing = stored_run
-            next [ identical!(existing, digest), true ] if existing
+            next settled(existing, digest) if settles?(existing, digest)
+            next [ progress(existing, digest), false, true ] if existing
 
             admit!
-            [ import(digest), false ]
+            [ import(digest), false, false ]
           end
         end
       rescue ActiveRecord::RecordNotUnique
         # A concurrent delivery committed this run, or an agent, evaluation or
         # scenario this import was creating, first. Reading again sees it.
         existing = stored_run
-        return [ identical!(existing, digest), true ] if existing
+        return settled(existing, digest) if settles?(existing, digest)
 
         attempts += 1
         retry if attempts < 2
@@ -265,6 +282,29 @@ module ActionAgent
       EvaluationReportExport.find(@payload["run_id"], account: @account)
     end
 
+    def running_delivery?
+      @payload["status"] == "running"
+    end
+
+    # Whether a delivery under a stored run's run_id leaves its results as
+    # they are. Only a running run this collector imported takes new ones.
+    def settles?(run, digest)
+      return false unless run
+      return true unless run.running? && run.imported?
+
+      run.external_report_digest == digest
+    end
+
+    # Returns `[run, true, false]` for a delivery that leaves the stored run as
+    # it is. Raises Conflict for a run the dashboard is executing, and for a
+    # different "complete" report under a run that has finished.
+    def settled(run, digest)
+      raise Conflict, "run_id names a run the dashboard is executing" if run.running? && !run.imported?
+      return [ run, true, false ] if running_delivery? || run.running?
+
+      [ identical!(run, digest), true, false ]
+    end
+
     # The stored run, when the payload is the report it was stored from or
     # the run's own export (EvaluationReportExport), which is how an
     # exported run posted back to its install is told from a different
@@ -286,26 +326,60 @@ module ActionAgent
       raise Denied, denial if denial.present?
     end
 
+    # Replaces a running run's results with the delivery's.
+    def progress(run, digest)
+      agent = find_or_create_agent
+      evaluation = find_or_initialize_evaluation(agent)
+      raise Conflict, "run_id already exists for another agent, suite or scope" unless evaluation.id == run.evaluation_id
+
+      scenarios = upsert_scenarios(evaluation)
+      run.scenario_results.destroy_all
+      run.update!(run_attributes(agent, digest))
+      store_results(run, scenarios, agent)
+    end
+
+    # A new run is dated by the time its report says it finished, when it
+    # names one (see #finished_at), so a run exported from one install and
+    # imported into another keeps its place in the run history.
     def import(digest)
       agent = find_or_create_agent
       evaluation = find_or_initialize_evaluation(agent)
       scenarios = upsert_scenarios(evaluation)
       run = evaluation.evaluation_runs.create!(
-        external_tenant: tenant_key,
-        external_run_id: @payload["run_id"],
+        run_attributes(agent, digest).merge(external_tenant: tenant_key, external_run_id: @payload["run_id"], created_at: finished_at)
+      )
+      store_results(run, scenarios, agent)
+    end
+
+    def run_attributes(agent, digest)
+      {
         external_report_digest: digest,
         agent_version: reported_release_version(agent),
-        status: :complete,
+        status: running_delivery? ? :running : :complete,
         selection: selection,
         scores: recorded_scores,
         samples_evaluated: results.size,
         samples_passed: results.count { |result| result["status"] == "passed" },
-        completed_at: Time.current
-      )
+        completed_at: running_delivery? ? nil : finished_at
+      }
+    end
+
+    # The report's `metadata.finished`, which EvaluationRun#to_report records
+    # as the run's completion time, or now when the report names none. A
+    # time later than now is read as now.
+    def finished_at
+      @finished_at ||= begin
+        now = Time.current
+        reported = metadata["finished"] && Time.iso8601(metadata["finished"])
+        reported && reported < now ? reported : now
+      end
+    end
+
+    def store_results(run, scenarios, agent)
       results.each { |result| persist(run, scenarios.fetch(result["scenario_key"]), result) }
       run.update!(scores: summarized_scores(run))
-      evaluation.touch
-      agent.update_columns(last_observed_at: run.completed_at, updated_at: Time.current)
+      run.evaluation.touch
+      agent.update_columns(last_observed_at: Time.current, updated_at: Time.current)
       run
     end
 
@@ -621,6 +695,9 @@ module ActionAgent
       object!(@payload, "payload")
       validate_json!(@payload)
       raise Invalid, "version must be 1" unless @payload["version"] == 1
+      unless @payload["status"].nil? || RUN_STATUSES.include?(@payload["status"])
+        raise Invalid, "status must be one of #{RUN_STATUSES.join(', ')}"
+      end
 
       %w[run_id source suite].each { |key| identifier!(@payload[key], key) }
       unless @payload["agent_name"].is_a?(String) && AGENT_NAME_PATTERN.match?(@payload["agent_name"]) && @payload["agent_name"].strip.length >= 2
@@ -634,6 +711,7 @@ module ActionAgent
         raise Invalid, "suite and scope name an evaluation longer than #{MAX_EVALUATION_NAME} characters"
       end
 
+      validate_finished!
       judge_trace_ids!(metadata["judge_trace_ids"])
       string!(report["judge"], "report.judge", 200)
       validate_release!
@@ -736,6 +814,21 @@ module ActionAgent
       string!(judge.dig("suggested_tool", "name"), "result.diagnosis.judge.suggested_tool.name", 200)
     end
 
+    def validate_finished!
+      finished = metadata["finished"]
+      return if finished.nil?
+      return if finished.is_a?(String) && finished.length <= 40 && iso8601?(finished)
+
+      raise Invalid, "report.metadata.finished must be an ISO 8601 time"
+    end
+
+    def iso8601?(value)
+      Time.iso8601(value)
+      true
+    rescue ArgumentError
+      false
+    end
+
     # The release the report says it evaluated: a digest the dashboard can
     # match to a version cut on deploy, and the deploy and label as text.
     def validate_release!

data/app/services/action_agent/github_client.rb

@@ -60,6 +60,10 @@ module ActionAgent
     # cannot see, which GitHub does not tell apart. Also 409, which GitHub
     # answers for the commits and trees of a repository with no commits.
     class NotFound < Error; end
+    # GitHub answered 403 to an organization membership lookup: the GitHub
+    # App lacks the organization permission Members (read), or the
+    # organization's SAML single sign-on has not authorized the token.
+    class MembershipForbidden < Error; end
     # GitHub minted no token because the installation was removed from its
     # account (404) or suspended (403). +reason+ is :removed or :suspended.
     class InstallationUnavailable < Error
@@ -323,14 +327,19 @@ module ActionAgent
 
     # The user's membership in +organization+ ({ "state" =>, "role" => }), or
     # nil when GitHub reports none.
+    #
+    # @raise [MembershipForbidden] when GitHub answers 403, as it does to a
+    #   GitHub App user token while the App lacks Members (read)
     def organization_membership(organization)
       raise Error, "#{organization.inspect} is not a GitHub login" unless organization.is_a?(String) && organization.match?(LOGIN)
 
       get("/user/memberships/orgs/#{organization}")
+    rescue NotFound
+      nil
     rescue Error => e
-      raise unless e.status.in?([ 403, 404 ])
+      raise unless e.status == 403
 
-      nil
+      raise MembershipForbidden.new("GitHub refused to show the membership in #{organization}: #{e.message}", status: 403)
     end
 
     # The repositories an installation token reaches, reduced like

data/app/services/action_agent/local_sandbox_claude_login.rb

@@ -108,7 +108,7 @@ module ActionAgent
       data = JSON.parse(File.read(path, 8192))
       status = data["status"].to_s
       result = { status: %w[starting awaiting_code submitted completed failed expired cancelled].include?(status) ? status : "failed" }
-      if status == "awaiting_code" && data["authorize_url"].to_s.match?(%r{\Ahttps://(?:claude\.ai|platform\.claude\.com)/oauth/authorize\?[^\s]+\z})
+      if status == "awaiting_code" && data["authorize_url"].to_s.match?(%r{\Ahttps://(?:claude\.ai|claude\.com|platform\.claude\.com)/(?:cai/)?oauth/authorize\?[^\s]+\z})
         result[:authorize_url] = data["authorize_url"]
       end
       result

data/app/services/action_agent/project_capabilities.rb

@@ -14,7 +14,8 @@ module ActionAgent
   #                      project's secrets and bootstrap reach it
   #   execution          agent execution is enabled
   #   github             a GitHub App or OAuth app is configured
-  #   github_connection  the owner connected GitHub
+  #   github_connection  the owner linked a usable GitHub App installation
+  #                      or connected GitHub through OAuth
   #   model_credentials  credentials for the model the project's agent
   #                      runs on (not blocking)
   #   browser            a browser for the project's sandbox, started and
@@ -24,7 +25,8 @@ module ActionAgent
     Item = Struct.new(:key, :label, :ok, :blocking, :detail, :fix, keyword_init: true)
 
     # @param owner [Object, nil] the caller's owner, for its credentials
-    # @param github_connected [Boolean]
+    # @param github_connected [Boolean] whether the owner has a usable GitHub
+    #   App installation or an OAuth connection
     # @param base_url [String] the engine's absolute mount URL, for callback URLs
     # @param environment [ActiveSupport::EnvironmentInquirer]
     def initialize(owner:, github_connected:, base_url:, environment: Rails.env)
@@ -137,9 +139,10 @@ module ActionAgent
     end
 
     def github_connection_item
+      fix = github[:mode] == "app" ? "Install the GitHub App in Settings → Integrations" : "Connect GitHub in Settings → Integrations"
       Item.new(key: "github_connection", label: "GitHub connection", ok: @github_connected, blocking: true,
         detail: @github_connected ? "Connected" : "GitHub is not connected",
-        fix: @github_connected ? nil : "Connect GitHub in Settings → Integrations")
+        fix: @github_connected ? nil : fix)
     end
 
     def default_model

data/app/views/action_agent/traces/_styles.html.erb

@@ -0,0 +1,45 @@
+<style>
+  .trace-console { color-scheme: light dark; --background: #f8fafc; --surface: #fff; --text: #172033; --muted: #475569; --border: #cbd5e1; --link: #3730a3; --danger: #b91c1c; --success: #166534; margin: 0; background: var(--background); color: var(--text); font: 16px/1.5 system-ui, sans-serif; }
+  .trace-console * { box-sizing: border-box; }
+  .trace-console header { background: var(--surface); border-bottom: 1px solid var(--border); }
+  .trace-console nav, .trace-console main { max-width: 1200px; margin: auto; padding: 1rem; }
+  .trace-console nav, .trace-console .refresh-controls, .trace-console .tokens { display: flex; flex-wrap: wrap; gap: 1rem; align-items: center; }
+  .trace-console a { color: var(--link); text-underline-offset: .2em; }
+  .trace-console a:focus-visible, .trace-console summary:focus-visible { outline: 3px solid var(--link); outline-offset: 3px; }
+  .trace-console .muted, .trace-console dt { color: var(--muted); }
+  .trace-console .refresh-controls { justify-content: flex-end; margin-bottom: 1rem; }
+  .trace-console h1 { font-size: 1.5rem; }
+  .trace-console h2 { font-size: 1.2rem; }
+  .trace-console .panel { background: var(--surface); border: 1px solid var(--border); border-radius: .5rem; padding: 1rem; margin: 1rem 0; }
+  .trace-console .summary, .trace-console .metadata { display: grid; grid-template-columns: repeat(auto-fit, minmax(160px, 1fr)); gap: 1rem; margin: 1rem 0; }
+  .trace-console .summary > div { padding: 1rem; background: var(--surface); border: 1px solid var(--border); border-radius: .5rem; }
+  .trace-console dt { font-size: .875rem; }
+  .trace-console dd { margin: 0; overflow-wrap: anywhere; }
+  .trace-console .summary dd { font-size: 1.4rem; font-weight: 600; }
+  .trace-console details { margin: .5rem 0; }
+  .trace-console summary { cursor: pointer; padding: .5rem 0; }
+  .trace-console .trace-summary { display: inline-flex; flex-wrap: wrap; gap: .5rem 1rem; align-items: baseline; }
+  .trace-console .badge { border: 1px solid currentColor; border-radius: .25rem; padding: .1rem .4rem; font-size: .8rem; }
+  .trace-console .error { color: var(--danger); }
+  .trace-console .success { color: var(--success); }
+  .trace-console .error-message { border-left: 4px solid var(--danger); padding: .5rem 1rem; overflow-wrap: anywhere; }
+  .trace-console pre { font: .8rem/1.5 ui-monospace, monospace; overflow: auto; padding: 1rem; background: var(--background); border: 1px solid var(--border); }
+  .trace-console .table-scroll { overflow-x: auto; }
+  .trace-console table { width: 100%; border-collapse: collapse; text-align: left; }
+  .trace-console th, .trace-console td { padding: .75rem; border-bottom: 1px solid var(--border); }
+  .trace-console .span-row { display: grid; grid-template-columns: minmax(5rem, 9rem) 1fr 5rem; gap: .5rem; align-items: center; margin: .5rem 0; font-size: .8rem; }
+  .trace-console .span-name { overflow: hidden; white-space: nowrap; text-overflow: ellipsis; }
+  .trace-console .span-track { position: relative; height: 1.25rem; background: var(--background); border: 1px solid var(--border); overflow: hidden; border-radius: .2rem; }
+  .trace-console .span-bar { position: absolute; height: 100%; background: #6366f1; }
+  .trace-console .span-bar[data-type="tool"] { background: #b45309; }
+  .trace-console .span-bar[data-type="thinking"] { background: #9333ea; }
+  .trace-console .span-bar[data-type="response"] { background: #15803d; }
+  .trace-console .span-bar[data-status="ERROR"] { background: var(--danger); }
+  @media (prefers-color-scheme: dark) {
+    .trace-console { --background: #111827; --surface: #1f2937; --text: #f1f5f9; --muted: #cbd5e1; --border: #64748b; --link: #c7d2fe; --danger: #fca5a5; --success: #86efac; }
+  }
+  @media (max-width: 480px) {
+    .trace-console main { padding: .5rem; }
+    .trace-console .span-row { grid-template-columns: 5rem 1fr 4rem; }
+  }
+</style>

data/app/views/action_agent/traces/_summary.html.erb

@@ -0,0 +1,8 @@
+<dl class="summary" aria-label="Last 24 hours">
+  <div><dt>Total Traces</dt><dd><%= number_with_delimiter(@metrics[:total_traces]) %></dd></div>
+  <div><dt>Total Tokens</dt><dd><%= number_to_human(@metrics[:total_tokens], precision: 2) %></dd></div>
+  <div><dt>Avg Duration</dt><dd><%= @metrics[:avg_duration_ms] >= 1000 ? "#{(@metrics[:avg_duration_ms] / 1000.0).round(2)}s" : "#{@metrics[:avg_duration_ms].round(0)}ms" %></dd></div>
+  <div><dt>Error Rate</dt><dd class="<%= 'error' if @metrics[:error_rate] > 5 %>"><%= @metrics[:error_rate] %>%</dd></div>
+  <div><dt>Active Agents</dt><dd><%= @metrics[:unique_agents] %></dd></div>
+</dl>
+<p class="muted">Last 24h</p>

data/app/views/action_agent/traces/_trace_detail.html.erb

@@ -1,64 +1,22 @@
-<div class="space-y-4">
-  <!-- Trace Info -->
-  <div class="grid grid-cols-2 md:grid-cols-4 gap-4 text-sm">
-    <div>
-      <span class="text-gray-500 dark:text-gray-400">Trace ID</span>
-      <div class="font-mono text-gray-900 dark:text-white"><%= trace.trace_id %></div>
-    </div>
-    <div>
-      <span class="text-gray-500 dark:text-gray-400">Service</span>
-      <div class="text-gray-900 dark:text-white"><%= trace.service_name || "—" %></div>
-    </div>
-    <div>
-      <span class="text-gray-500 dark:text-gray-400">Provider</span>
-      <div class="text-gray-900 dark:text-white"><%= trace.provider || "—" %></div>
-    </div>
-    <div>
-      <span class="text-gray-500 dark:text-gray-400">Model</span>
-      <div class="text-gray-900 dark:text-white"><%= trace.model || "—" %></div>
-    </div>
-  </div>
-
-  <% if trace.error? && trace.error_message.present? %>
-    <div class="bg-red-50 dark:bg-red-900/20 border border-red-200 dark:border-red-800 rounded p-3">
-      <div class="flex">
-        <svg class="h-5 w-5 text-red-400" viewBox="0 0 20 20" fill="currentColor">
-          <path fill-rule="evenodd" d="M10 18a8 8 0 100-16 8 8 0 000 16zM8.707 7.293a1 1 0 00-1.414 1.414L8.586 10l-1.293 1.293a1 1 0 101.414 1.414L10 11.414l1.293 1.293a1 1 0 001.414-1.414L11.414 10l1.293-1.293a1 1 0 00-1.414-1.414L10 8.586 8.707 7.293z" clip-rule="evenodd" />
-        </svg>
-        <div class="ml-3">
-          <h3 class="text-sm font-medium text-red-800 dark:text-red-200">Error</h3>
-          <p class="mt-1 text-sm text-red-700 dark:text-red-300"><%= trace.error_message %></p>
-        </div>
-      </div>
-    </div>
+<dl class="metadata">
+  <div><dt>Trace ID</dt><dd><%= trace.trace_id %></dd></div>
+  <div><dt>Service</dt><dd><%= trace.service_name || "Unknown" %></dd></div>
+  <div><dt>Provider</dt><dd><%= trace.provider || "Unknown" %></dd></div>
+  <div><dt>Model</dt><dd><%= trace.model || "Unknown" %></dd></div>
+</dl>
+<% if trace.error? && trace.error_message.present? %>
+  <p class="error-message error"><strong>Error:</strong> <%= trace.error_message %></p>
+<% end %>
+<div class="tokens">
+  <span>Input: <%= number_with_delimiter(trace.total_input_tokens || 0) %></span>
+  <span>Output: <%= number_with_delimiter(trace.total_output_tokens || 0) %></span>
+  <% if trace.total_thinking_tokens.to_i > 0 %>
+    <span>Thinking: <%= number_with_delimiter(trace.total_thinking_tokens) %></span>
   <% end %>
-
-  <!-- Token Breakdown -->
-  <div class="flex gap-6 text-sm">
-    <div class="flex items-center gap-2">
-      <div class="w-3 h-3 rounded-full bg-blue-500"></div>
-      <span class="text-gray-500">Input:</span>
-      <span class="font-medium text-gray-900 dark:text-white"><%= number_with_delimiter(trace.total_input_tokens || 0) %></span>
-    </div>
-    <div class="flex items-center gap-2">
-      <div class="w-3 h-3 rounded-full bg-green-500"></div>
-      <span class="text-gray-500">Output:</span>
-      <span class="font-medium text-gray-900 dark:text-white"><%= number_with_delimiter(trace.total_output_tokens || 0) %></span>
-    </div>
-    <% if trace.total_thinking_tokens.to_i > 0 %>
-      <div class="flex items-center gap-2">
-        <div class="w-3 h-3 rounded-full bg-purple-500"></div>
-        <span class="text-gray-500">Thinking:</span>
-        <span class="font-medium text-purple-600 dark:text-purple-400"><%= number_with_delimiter(trace.total_thinking_tokens) %></span>
-      </div>
-    <% end %>
-  </div>
-
-  <!-- Spans Timeline -->
-  <% if trace.spans.present? %>
-    <div class="mt-4">
-      <h4 class="text-sm font-medium text-gray-700 dark:text-gray-300 mb-2">Span Timeline</h4>
-      <div class="relative bg-gray-100 dark:bg-gray-800 rounded-lg p-4">
+</div>
+<% if trace.spans.present? %>
+  <section aria-label="Span timeline">
+    <h2>Span Timeline</h2>
         <% total_duration = (trace.total_duration_ms || 1).to_f %>
         <% total_duration = 1.0 if total_duration.zero? %>
         <%
@@ -77,41 +35,20 @@
             start_pct = ((relative_ms || 0).to_f / total_duration * 100).clamp(0, 100)
             width_pct = ((span["duration_ms"] || 0).to_f / total_duration * 100).clamp(0.5, [ 100 - start_pct, 0.5 ].max)
             span_type = span["type"] || "unknown"
-            colors = {
-              "root" => "bg-gray-400",
-              "prompt" => "bg-blue-400",
-              "llm" => "bg-indigo-500",
-              "generate" => "bg-indigo-400",
-              "tool" => "bg-amber-500",
-              "thinking" => "bg-purple-500",
-              "response" => "bg-green-400"
-            }
-            bg_color = colors[span_type] || "bg-gray-400"
           %>
-          <div class="flex items-center gap-2 mb-2">
-            <div class="w-24 text-xs text-gray-500 dark:text-gray-400 truncate" title="<%= span["name"] %>">
-              <%= span["name"]&.split(".")&.last || span_type %>
-            </div>
-            <div class="flex-1 h-6 bg-gray-200 dark:bg-gray-700 rounded relative overflow-hidden">
-              <div class="span-bar absolute h-full <%= bg_color %> rounded <%= span["status"] == "ERROR" ? 'opacity-50 bg-red-500' : '' %>"
-                   style="left: <%= start_pct %>%; width: <%= width_pct %>%;"
-                   title="<%= span["name"] %>: <%= span["duration_ms"]&.round(1) %>ms">
-              </div>
-            </div>
-            <div class="w-16 text-xs text-gray-500 dark:text-gray-400 text-right">
-              <%= span["duration_ms"] ? "#{span["duration_ms"].round(1)}ms" : "—" %>
-            </div>
-          </div>
-        <% end %>
+      <div class="span-row">
+        <span class="span-name" title="<%= span["name"] %>"><%= span["name"]&.split(".")&.last || span_type %></span>
+        <div class="span-track">
+          <div class="span-bar" data-type="<%= span_type %>" data-status="<%= span["status"] %>"
+               style="left: <%= start_pct %>%; width: <%= width_pct %>%;"
+               title="<%= span["name"] %>: <%= span["duration_ms"]&.round(1) %>ms"></div>
+        </div>
+        <span><%= span["duration_ms"] ? "#{span["duration_ms"].round(1)}ms" : "Unknown" %></span>
       </div>
-    </div>
-  <% end %>
-
-  <!-- Raw Data (collapsed) -->
-  <details class="mt-4">
-    <summary class="text-sm text-gray-500 dark:text-gray-400 cursor-pointer hover:text-gray-700">
-      View raw trace data
-    </summary>
-    <pre class="mt-2 p-3 bg-gray-100 dark:bg-gray-800 rounded text-xs overflow-x-auto"><%= JSON.pretty_generate(trace.as_json(except: [:id, :created_at, :updated_at])) %></pre>
-  </details>
-</div>
+    <% end %>
+  </section>
+<% end %>
+<details>
+  <summary>View raw trace data</summary>
+  <pre><%= JSON.pretty_generate(trace.as_json(except: [ :id, :created_at, :updated_at ])) %></pre>
+</details>