feat: stricter liquid rendering

enterprise/app/models/concerns/toolable.rb

@@ -63,7 +63,11 @@ module Concerns::Toolable
   private
 
   def render_template(template, context)
-    Liquid::Template.parse(template).render(context.deep_stringify_keys)
+    liquid_template = Liquid::Template.parse(template, error_mode: :strict)
+    liquid_template.render(context.deep_stringify_keys, registers: {}, strict_variables: true, strict_filters: true)
+  rescue Liquid::SyntaxError, Liquid::UndefinedVariable, Liquid::UndefinedFilter => e
+    Rails.logger.error("Liquid template error: #{e.message}")
+    raise "Template rendering failed: #{e.message}"
   end
 
   def parse_response_body(body)

enterprise/lib/captain/tools/http_tool.rb

@@ -35,7 +35,9 @@ class Captain::Tools::HttpTool < Agents::Tool
     IPAddr.new('fe80::/10')       # IPv6 Link-local
   ].freeze
 
-  MAX_RESPONSE_SIZE = 2.megabytes
+  # Limit response size to prevent memory exhaustion and match LLM token limits
+  # 1MB of text ≈ 250K tokens, which exceeds most LLM context windows
+  MAX_RESPONSE_SIZE = 1.megabyte
 
   def execute_http_request(url, body)
     uri = URI.parse(url)