actionagent 1.9.1 → 1.9.3

data/lib/action_agent.rb

@@ -111,6 +111,9 @@ module ActionAgent
   # What ActionAgent.claude_code_auth may be set to.
   CLAUDE_CODE_AUTH_MODES = %i[api_key local_login sandbox_login].freeze
 
+  # What ActionAgent.codex_auth may be set to.
+  CODEX_AUTH_MODES = %i[api_key local_login].freeze
+
   # What ActionAgent.provider_key_scope may be set to.
   PROVIDER_KEY_SCOPES = %i[organization personal_override].freeze
 
@@ -134,6 +137,9 @@ module ActionAgent
   #                           they are handed to, or delete a project that has
   #                           them
   #   :take_over_browser      drive a run's browser by hand
+  #   :chat_code_session      send a message to a Claude Code or Codex
+  #                           conversation someone may have started (the
+  #                           subject is its first session)
   #   :manage_recordings      delete a session recording
   #   :replace_scenarios      replace or merge an evaluation's scenarios
   PERMISSION_ACTIONS = %i[
@@ -144,6 +150,7 @@ module ActionAgent
     answer_input_request
     manage_project_secrets
     take_over_browser
+    chat_code_session
     manage_recordings
     replace_scenarios
   ].freeze
@@ -589,10 +596,35 @@ module ActionAgent
     # @return [Integer]
     attr_accessor :claude_code_timeout
 
-    # Codex CLI sessions use the owner's OpenAI API key and workspace-write
-    # sandboxing. They share checkout lifecycle and cancellation with Claude.
+    # Codex CLI sessions run with workspace-write sandboxing, and share the
+    # checkout lifecycle and cancellation with Claude Code.
     attr_accessor :codex_command, :codex_timeout
 
+    # How Codex sessions authenticate.
+    #
+    # :api_key (the default) runs them on the OpenAI API key the owner
+    # connected in Settings -> Integrations, handed to the session as
+    # CODEX_API_KEY with a Codex home of the sandbox's own. Usage is billed
+    # to that key's project.
+    #
+    # :local_login runs `codex` on whatever login this machine's user set up
+    # with `codex login` (a ChatGPT plan, or a key given to the CLI), which
+    # Codex keeps under ~/.codex. The dashboard never reads, copies or
+    # stores it; it only asks `codex login status` whether there is one.
+    # That login is the dashboard user's own, so this works with the :local
+    # sandbox backend only, and other backends refuse Codex sessions.
+    # @return [Symbol] :api_key or :local_login
+    attr_reader :codex_auth
+
+    def codex_auth=(value)
+      mode = value.to_s.to_sym
+      unless CODEX_AUTH_MODES.include?(mode)
+        raise ArgumentError, "ActionAgent.codex_auth must be :api_key or :local_login, not #{value.inspect}"
+      end
+
+      @codex_auth = mode
+    end
+
     # How Claude Code sessions authenticate.
     #
     # :api_key (the default) runs them on the Anthropic API key the owner
@@ -685,6 +717,20 @@ module ActionAgent
     # @return [Proc, nil]
     attr_accessor :scenario_evaluation_adapter_resolver
 
+    # Whether a scenario evaluation runs as background jobs, one per
+    # scenario x model (EvaluationReplayJob) followed by one to score it
+    # (EvaluationJudgeJob), instead of as one job that replays the suite in
+    # order. The jobs run on the :agents queue, so a run is as parallel as
+    # the host's workers for that queue: a single-worker box stays
+    # sequential. The last result to be scored completes the run.
+    #
+    # A run that drives a sandbox's browser, and a run a host adapter
+    # (scenario_evaluation_adapter_resolver) replays, stay sequential: the
+    # browser is one page, and the adapter owns its own scheduling.
+    # Off unless set.
+    # @return [Boolean]
+    attr_accessor :scenario_evaluation_fan_out
+
     # How many answerable candidates an exploration's review pre-selects for
     # acceptance: an Integer, or a Proc called with (owner) that returns one.
     # A platform can cap a free plan's suite this way. Unset, or nil from the
@@ -1234,6 +1280,7 @@ module ActionAgent
       @claude_code_timeout = 1800
       @codex_command = "codex"
       @codex_timeout = 1800
+      @codex_auth = :api_key
       @claude_code_auth = :api_key
       @claude_code_hosted_login_enabled = false
       @claude_code_login_timeout = 300
@@ -1243,6 +1290,7 @@ module ActionAgent
       @capture_dashboard_sessions = true
 
       @scenario_evaluation_adapter_resolver = nil
+      @scenario_evaluation_fan_out = false
       @table_name_prefix = "active_agent_"
       @agent_polymorphic_name = nil
       @encrypt_credentials = true

data/lib/generators/action_agent/templates/migrations/028_add_conversations_to_code_sessions.rb.erb

@@ -0,0 +1,18 @@
+# frozen_string_literal: true
+
+# Makes Claude Code and Codex sessions conversations: each session is one
+# turn, and a turn after the first names the conversation's first session as
+# its root. A conversation resumes the same CLI session (resume_session_id),
+# and a turn may carry a note from the dashboard ahead of the person's
+# message (context_note), such as who drove the browser since the last turn.
+class AddConversationsToCodeSessions < ActiveRecord::Migration<%= migration_version %>
+  def change
+    sessions = "#{ActionAgent.table_name_prefix}code_sessions"
+    add_column sessions, :conversation_root_id, :bigint
+    add_column sessions, :turn_index, :integer, default: 0, null: false
+    add_column sessions, :origin, :string
+    add_column sessions, :context_note, :text
+    add_column sessions, :resume_session_id, :string
+    add_index sessions, :conversation_root_id
+  end
+end

data/lib/tasks/evaluations.rake

@@ -0,0 +1,78 @@
+# frozen_string_literal: true
+
+# Evaluation run reports from the command line: a run out as the version-1
+# envelope the evaluation report collector takes (or as its HTML report page,
+# or as GitHub-flavored Markdown for a pull request), and an envelope in
+# through the collector's own import. On a multi-tenant install name the
+# tenant with ACCOUNT_ID=<id>.
+#
+#   bin/rails action_agent:evaluations:export RUN=dashboard-run-12-20261021093000 FILE=tmp/run.json
+#   bin/rails action_agent:evaluations:export EVALUATION=3 RUN=12 FORMAT=html FILE=tmp/run.html
+#   bin/rails action_agent:evaluations:export EVALUATION=3 RUN=12 FORMAT=md > tmp/run.md
+#   bin/rails action_agent:evaluations:import FILE=tmp/run.json
+namespace :action_agent do
+  namespace :evaluations do
+    desc "Write a run's report (RUN=run_id, or EVALUATION=id RUN=id; FORMAT=json|html|md; FILE=path, standard output without FILE)"
+    task export: :environment do
+      run = action_agent_evaluation_run!
+      export = ActionAgent::EvaluationReportExport.new(run)
+      format = ENV.fetch("FORMAT", "json").downcase
+      body =
+        case format
+        when "json" then export.to_json
+        when "html" then run.to_report.to_html
+        when "md", "markdown" then export.markdown
+        else abort "FORMAT is json, html or md"
+        end
+
+      if ENV["FILE"].present?
+        File.write(ENV["FILE"], body)
+        puts "Wrote run #{export.run_id} of #{run.evaluation.name} (#{format}) to #{ENV['FILE']}"
+      else
+        puts body
+      end
+    rescue ActionAgent::EvaluationReportExport::NotExportable => e
+      abort "Export refused: #{e.message}"
+    end
+
+    desc "Store a version-1 evaluation report envelope (FILE=path [ACCOUNT_ID=id]); importing it again changes nothing"
+    task import: :environment do
+      path = ENV.fetch("FILE") { abort "Name the file: FILE=tmp/run.json" }
+      abort "No file at #{path}" unless File.file?(path)
+      abort "Unavailable: #{ActionAgent::EvaluationReportImport.unavailable_reason}" if ActionAgent::EvaluationReportImport.unavailable_reason
+
+      account = action_agent_evaluation_account
+      run, duplicate = ActionAgent::EvaluationReportImport.call(account: account, payload: JSON.parse(File.read(path)))
+      ActionAgent.record_usage(account, :evaluation_report) unless duplicate
+      puts "#{duplicate ? 'Already stored (identical retry)' : 'Stored'}: run #{ActionAgent::EvaluationReportExport.run_id_for(run)} " \
+           "as run ##{run.id} of evaluation ##{run.evaluation_id} (#{run.evaluation.name}), #{run.status}"
+    rescue JSON::ParserError
+      abort "Import refused: #{path} is not JSON"
+    rescue ActionAgent::EvaluationReportImport::Error, ActiveRecord::RecordInvalid => e
+      abort "Import refused: #{e.message}"
+    end
+  end
+end
+
+# The tenant, from ACCOUNT_ID (ACCOUNT also works); nil on a single-tenant
+# install.
+def action_agent_evaluation_account
+  id = ENV["ACCOUNT_ID"].presence || ENV["ACCOUNT"].presence
+  return nil unless id
+
+  abort "ACCOUNT_ID names a tenant, which only a multi-tenant install has" unless ActionAgent.multi_tenant?
+  ActionAgent.account_class.to_s.constantize.find_by(id: id) || abort("No account #{id}")
+end
+
+# The run RUN names: with EVALUATION, a run id of that evaluation; without,
+# a run_id as the export endpoint takes it, within the tenant.
+def action_agent_evaluation_run!
+  run_id = ENV.fetch("RUN") { abort "Name the run: RUN=<run_id>, or EVALUATION=<id> RUN=<id>" }
+  if ENV["EVALUATION"].present?
+    evaluation = ActionAgent::Evaluation.find_by(id: ENV["EVALUATION"]) || abort("No evaluation #{ENV['EVALUATION']}")
+    evaluation.evaluation_runs.find_by(id: run_id) || abort("No run #{run_id} in evaluation #{evaluation.id}")
+  else
+    ActionAgent::EvaluationReportExport.find(run_id, account: action_agent_evaluation_account) ||
+      abort("No run with run_id #{run_id.inspect}#{' for that account' if ENV['ACCOUNT_ID'].present? || ENV['ACCOUNT'].present?}")
+  end
+end

metadata

@@ -1,14 +1,14 @@
 --- !ruby/object:Gem::Specification
 name: actionagent
 version: !ruby/object:Gem::Version
-  version: 1.9.1
+  version: 1.9.3
 platform: ruby
 authors:
 - Justin Bowen
 autorequire:
 bindir: bin
 cert_chain: []
-date: 2026-10-17 00:00:00.000000000 Z
+date: 2026-10-21 00:00:00.000000000 Z
 dependencies:
 - !ruby/object:Gem::Dependency
   name: activeagent
@@ -190,6 +190,8 @@ files:
 - app/jobs/action_agent/application_job.rb
 - app/jobs/action_agent/code_session_job.rb
 - app/jobs/action_agent/draft_pull_request_job.rb
+- app/jobs/action_agent/evaluation_judge_job.rb
+- app/jobs/action_agent/evaluation_replay_job.rb
 - app/jobs/action_agent/evaluation_run_job.rb
 - app/jobs/action_agent/exploration_job.rb
 - app/jobs/action_agent/input_request_expiry_job.rb
@@ -244,7 +246,9 @@ files:
 - app/models/concerns/action_agent/session_recordable.rb
 - app/queries/action_agent/agent_executions.rb
 - app/queries/action_agent/metrics_report.rb
+- app/queries/action_agent/trace_window_report.rb
 - app/serializers/action_agent/agent_message_serializer.rb
+- app/serializers/action_agent/code_conversation_serializer.rb
 - app/serializers/action_agent/evaluation_serializer.rb
 - app/serializers/action_agent/input_request_serializer.rb
 - app/serializers/action_agent/interaction_preview.rb
@@ -259,17 +263,23 @@ files:
 - app/services/action_agent/agent_sync.rb
 - app/services/action_agent/agent_tool_roster.rb
 - app/services/action_agent/agent_toolbox.rb
+- app/services/action_agent/browser_client_token.rb
 - app/services/action_agent/browser_live_ticket.rb
 - app/services/action_agent/browser_sidecar.rb
 - app/services/action_agent/browser_sign_in.rb
 - app/services/action_agent/browser_tool_redaction.rb
+- app/services/action_agent/checkout_tools.rb
 - app/services/action_agent/claude_code_auth.rb
+- app/services/action_agent/code_conversation.rb
+- app/services/action_agent/codex_auth.rb
 - app/services/action_agent/codex_session_events.rb
 - app/services/action_agent/dashboard_assistant_service.rb
+- app/services/action_agent/diff_stat.rb
 - app/services/action_agent/draft_pull_request_publisher.rb
 - app/services/action_agent/evaluation_evidence.rb
 - app/services/action_agent/evaluation_fix.rb
 - app/services/action_agent/evaluation_fix_comparison.rb
+- app/services/action_agent/evaluation_report_export.rb
 - app/services/action_agent/evaluation_report_import.rb
 - app/services/action_agent/evaluation_run_cost.rb
 - app/services/action_agent/evaluation_runner_service.rb
@@ -314,6 +324,7 @@ files:
 - app/services/action_agent/session_index.rb
 - app/services/action_agent/session_recording_service.rb
 - app/services/action_agent/session_timeline.rb
+- app/services/action_agent/slack_tools.rb
 - app/services/action_agent/tool_discovery.rb
 - app/views/action_agent/dashboard/index.html.erb
 - app/views/action_agent/github_app_manifests/created.html.erb
@@ -358,9 +369,11 @@ files:
 - lib/generators/action_agent/templates/migrations/025_create_active_agent_scenario_catalogs.rb.erb
 - lib/generators/action_agent/templates/migrations/026_add_evaluation_fixes_to_code_sessions.rb.erb
 - lib/generators/action_agent/templates/migrations/027_create_active_agent_interaction_feedbacks.rb.erb
+- lib/generators/action_agent/templates/migrations/028_add_conversations_to_code_sessions.rb.erb
 - lib/tasks/action_agent.rake
 - lib/tasks/browser.rake
 - lib/tasks/claude_code.rake
+- lib/tasks/evaluations.rake
 - lib/tasks/sandbox.rake
 - lib/tasks/scenario_catalogs.rake
 homepage: https://activeagents.ai