Fix agent source registration and retire legacy Gemini assessment pipeline

.env.example

@@ -1 +1,5 @@
 GEMINI_API_KEY=
+
+# Hosted defuddle worker (content extraction). URL defaults to the production worker.
+DEFUDDLE_API_KEY=
+DEFUDDLE_API_URL=https://deffudler.svc.canadasbuilding.com

app/controllers/api/agent/commitment_matches_controller.rb

@@ -8,10 +8,13 @@ def create
           matchable_id: params.require(:matchable_id),
         )
 
+        # A match created by the agent has, by definition, already been reviewed.
         match.assign_attributes(
           relevance_score: params.require(:relevance_score),
           relevance_reasoning: params[:relevance_reasoning],
           matched_at: Time.current,
+          assessed: true,
+          assessed_at: Time.current,
         )
 
         match.save!

app/controllers/api/agent/pages_controller.rb

@@ -1,22 +1,24 @@
 module Api
   module Agent
     class PagesController < BaseController
-      ALLOWED_SUFFIXES = %w[.canada.ca .gc.ca].freeze
+      ALLOWED_SUFFIXES = %w[.canada.ca .gc.ca .parl.ca].freeze
 
       def fetch
         url = params.require(:url)
         government_id = params.require(:government_id)
 
         unless government_url?(url)
-          render json: { error: "URL must be on a *.canada.ca or *.gc.ca domain" }, status: :unprocessable_entity
+          render json: { error: "URL must be on a *.canada.ca, *.gc.ca, or *.parl.ca domain" }, status: :unprocessable_entity
           return
         end
 
         # Fetch and parse the page
-        response = HTTP.timeout(connect: 5, read: 20)
-          .headers("User-Agent" => "BuildCanada-Tracker/1.0")
-          .follow(max_hops: 3)
-          .get(url)
+        begin
+          response = PageFetcher.get(url)
+        rescue HTTP::TimeoutError, HTTP::ConnectionError, OpenSSL::SSL::SSLError, SocketError => e
+          render json: { error: "Failed to fetch: #{e.class.name.demodulize} (#{e.message})" }, status: :bad_gateway
+          return
+        end
 
         unless response.status.success?
           render json: { error: "Failed to fetch: HTTP #{response.status}" }, status: :bad_gateway
@@ -31,8 +33,13 @@ def fetch
         end
 
         # Parse HTML to markdown
-        prepared_html = Defuddle.prepare_html(response.body.to_s)
-        parsed_markdown, _parsed_html = Defuddle.defuddle(prepared_html)
+        begin
+          prepared_html = Defuddle.prepare_html(response.body.to_s)
+          parsed_markdown, _parsed_html = Defuddle.defuddle(prepared_html, url: final_url)
+        rescue Defuddle::ParseError => e
+          render json: { error: "Failed to parse page: #{e.message}" }, status: :bad_gateway
+          return
+        end
 
         # Extract title and date from meta tags
         doc = Nokogiri::HTML(response.body.to_s)

app/controllers/statcan_datasets_controller.rb

@@ -1,6 +1,11 @@
 class StatcanDatasetsController < ApplicationController
   def show
-    dataset = StatcanDataset.find_by(name: params[:id])
-    render json: dataset
+    dataset = StatcanDataset.find_by(name: params[:id]) || StatcanDataset.find_by(id: params[:id])
+
+    if dataset
+      render json: dataset
+    else
+      render json: { error: "Dataset not found" }, status: :not_found
+    end
   end
 end

app/jobs/agent_evaluate_commitment_job.rb

@@ -59,7 +59,7 @@ def build_cmd(prompt, hook_script:)
       "--system-prompt", system_prompt,
       "--allowedTools", allowed_tools.join(","),
       "--permission-mode", "bypassPermissions",
-      "--model", ENV.fetch("AGENT_MODEL", "claude-sonnet-4-6"),
+      "--model", ENV.fetch("AGENT_MODEL", "claude-sonnet-5"),
       "--output-format", "text",
       "--settings", hook_settings
     ]
@@ -91,7 +91,7 @@ def agent_env(commitment_id: nil, entry_id: nil)
       "CLAUDE_CODE_OAUTH_TOKEN" => ENV["CLAUDE_CODE_OAUTH_TOKEN"],
       "RAILS_API_URL"         => ENV.fetch("RAILS_API_URL", "http://localhost:3000"),
       "RAILS_API_KEY"         => Rails.application.credentials.dig(:agent, :api_key) || ENV["AGENT_API_KEY"],
-      "AGENT_MODEL"           => ENV.fetch("AGENT_MODEL", "claude-sonnet-4-6"),
+      "AGENT_MODEL"           => ENV.fetch("AGENT_MODEL", "claude-sonnet-5"),
       "COMMITMENT_ID"         => commitment_id&.to_s,
       "ENTRY_ID"              => entry_id&.to_s,
       # Explicitly unset — subprocess must not access Rails credentials

app/jobs/agent_process_entry_job.rb

@@ -48,7 +48,7 @@ def build_cmd(prompt, hook_script:)
       "--system-prompt", system_prompt,
       "--allowedTools", allowed_tools.join(","),
       "--permission-mode", "bypassPermissions",
-      "--model", ENV.fetch("AGENT_MODEL", "claude-opus-4-6"),
+      "--model", ENV.fetch("AGENT_MODEL", "claude-sonnet-5"),
       "--output-format", "text",
       "--settings", hook_settings
     ]
@@ -79,7 +79,7 @@ def agent_env(commitment_id: nil, entry_id: nil)
       "CLAUDE_CODE_OAUTH_TOKEN" => ENV["CLAUDE_CODE_OAUTH_TOKEN"],
       "RAILS_API_URL"     => ENV.fetch("RAILS_API_URL", "http://localhost:3000"),
       "RAILS_API_KEY"     => Rails.application.credentials.dig(:agent, :api_key) || ENV["AGENT_API_KEY"],
-      "AGENT_MODEL"       => ENV.fetch("AGENT_MODEL", "claude-opus-4-6"),
+      "AGENT_MODEL"       => ENV.fetch("AGENT_MODEL", "claude-sonnet-5"),
       "COMMITMENT_ID"        => commitment_id&.to_s,
       "ENTRY_ID"             => entry_id&.to_s,
       # Explicitly unset — subprocess must not access Rails credentials

app/models/chat.rb

@@ -3,6 +3,16 @@ class Chat < ApplicationRecord
 
   belongs_to :record, polymorphic: true, optional: true
 
+  class << self
+    # Rows from retired pipelines (CriterionAssessor, CommitmentStatusDeriver) remain as an
+    # audit log after their classes were removed. Load them as plain Chats instead of raising.
+    def find_sti_class(type_name)
+      super
+    rescue ActiveRecord::SubclassNotFound
+      self
+    end
+  end
+
   def system_prompt
   end