lex-llm 0.4.15 → 0.4.18

data/lib/legion/extensions/llm/connection.rb

@@ -42,6 +42,7 @@ module Legion
 
           ensure_configured!
           @connection ||= Faraday.new(provider.api_base) do |faraday|
+            faraday.ssl.verify = false
             setup_timeout(faraday)
             setup_logging(faraday)
             setup_retry(faraday)
@@ -82,6 +83,7 @@ module Legion
                            errors: false,
                            headers: false,
                            log_level: :debug do |logger|
+            logger.filter(logging_regexp('Bearer [A-Za-z0-9._\\-]+'), 'Bearer [REDACTED]')
             logger.filter(logging_regexp('[A-Za-z0-9+/=]{100,}'), '[BASE64 DATA]')
             logger.filter(logging_regexp('[-\\d.e,\\s]{100,}'), '[EMBEDDINGS ARRAY]')
           end
@@ -121,7 +123,13 @@ module Legion
           faraday.request :multipart
           faraday.request :json
           faraday.response :json
-          faraday.adapter :net_http
+          # Prefer typhoeus (libcurl) over net_http to avoid Ruby 4.0 + net-http-0.9.1 SSL issues
+          begin
+            require 'faraday/typhoeus'
+            faraday.adapter :typhoeus
+          rescue LoadError
+            faraday.adapter :net_http
+          end
           faraday.use :llm_errors, provider: @provider
         end
 
@@ -137,6 +145,7 @@ module Legion
             Timeout::Error,
             Faraday::TimeoutError,
             Faraday::ConnectionFailed,
+            Faraday::SSLError,
             Faraday::RetriableResponse,
             Legion::Extensions::Llm::RateLimitError,
             Legion::Extensions::Llm::ServerError,

data/lib/legion/extensions/llm/credential_sources.rb

@@ -18,6 +18,12 @@ module Legion
         CLAUDE_PROJECT  = File.join(Dir.pwd, '.claude', 'settings.json')
         CODEX_AUTH      = File.expand_path('~/.codex/auth.json')
 
+        def credential_source_probing_enabled?
+          return true unless defined?(::Legion::Settings)
+
+          ::Legion::Settings.dig(:extensions, :llm, :security, :credential_source_probing) != false
+        end
+
         # --- public helpers ------------------------------------------------
 
         # Fetch an environment variable, stripping whitespace.
@@ -30,9 +36,9 @@ module Legion
           stripped.empty? ? nil : stripped
         end
 
-        # Merged Claude config (user-level + project-level).  Project settings
-        # override user settings.  Memoized for the lifetime of the process.
         def claude_config
+          return {} unless credential_source_probing_enabled?
+
           @claude_config ||= merge_claude_configs
         end
 
@@ -52,9 +58,9 @@ module Legion
           env_hash[key.to_sym] || env_hash[key.to_s]
         end
 
-        # Read the bearer token from ~/.codex/auth.json when auth_mode is
-        # "chatgpt" and the JWT is not expired.
         def codex_token
+          return nil unless credential_source_probing_enabled?
+
           data = read_json(CODEX_AUTH)
           mode = data[:auth_mode] || data['auth_mode']
           return nil unless mode == 'chatgpt'
@@ -66,8 +72,9 @@ module Legion
           token
         end
 
-        # Read the OPENAI_API_KEY from ~/.codex/auth.json.
         def codex_openai_key
+          return nil unless credential_source_probing_enabled?
+
           data = read_json(CODEX_AUTH)
           val = data[:OPENAI_API_KEY] || data['OPENAI_API_KEY']
           return nil if val.nil?
@@ -206,7 +213,8 @@ module Legion
           false
         end
 
-        module_function :env, :claude_config, :claude_config_value,
+        module_function :env, :credential_source_probing_enabled?,
+                        :claude_config, :claude_config_value,
                         :claude_env_value, :codex_token, :codex_openai_key,
                         :setting, :socket_open?, :http_ok?,
                         :dedup_credentials, :credential_hash,

data/lib/legion/extensions/llm/fleet/token_validator.rb

@@ -28,7 +28,7 @@ module Legion
                                       verification_key: signing_key,
                                       issuer: issuer,
                                       algorithm: algorithm,
-                                      verify_issuer: false
+                                      verify_issuer: true
                                     ))
             validate_registered_claims!(claims)
             validate_request_expiry!(claims)
@@ -72,15 +72,30 @@ module Legion
             raise TokenError, 'fleet request expires_at is invalid'
           end
 
+          SCALAR_CLAIMS = %i[
+            request_id correlation_id idempotency_key operation provider provider_instance
+            model reply_to timeout_seconds expires_at
+          ].freeze
+          HASHABLE_CLAIMS = %i[message_context params caller trace_context].freeze
+
           def validate_envelope_claims!(claims, envelope)
-            %i[
-              request_id correlation_id idempotency_key operation provider provider_instance
-              model reply_to message_context params caller trace_context timeout_seconds expires_at
-            ].each do |key|
+            SCALAR_CLAIMS.each do |key|
               expected = canonical_value(envelope[key])
               actual = canonical_value(claims[key])
               raise TokenError, "fleet token #{key} claim mismatch" unless actual == expected
             end
+
+            HASHABLE_CLAIMS.each do |key|
+              hash_key = :"#{key}_hash"
+              expected_hash = content_hash(envelope[key])
+              actual_hash = claims[hash_key] || content_hash(claims[key])
+              raise TokenError, "fleet token #{key} hash mismatch" unless actual_hash == expected_hash
+            end
+          end
+
+          def content_hash(value)
+            require 'digest'
+            Digest::SHA256.hexdigest(canonical_value(value).to_s)
           end
 
           def reserve_replay!(jti)
@@ -119,8 +134,16 @@ module Legion
             @replay_mutex.synchronize { purge_replay_cache_locked!(Time.now.to_i) }
           end
 
+          MAX_REPLAY_ENTRIES = 100_000
+
           def purge_replay_cache_locked!(now)
             @seen_jtis.each_pair { |jti, entry| @seen_jtis.delete(jti) unless active_replay?(entry, now) }
+            evict_oldest_replay_entries! if @seen_jtis.size > MAX_REPLAY_ENTRIES
+          end
+
+          def evict_oldest_replay_entries!
+            sorted = @seen_jtis.each_pair.sort_by { |_jti, entry| entry[:expires_at] }
+            sorted.first(@seen_jtis.size - MAX_REPLAY_ENTRIES).each_key { |jti| @seen_jtis.delete(jti) }
           end
 
           def active_replay?(entry, now)

data/lib/legion/extensions/llm/fleet/worker_execution.rb

@@ -49,10 +49,11 @@ module Legion
             TokenValidator.validate!(token: envelope_value(envelope, :signed_token), envelope: envelope)
           end
 
-          def validate_policy!(_envelope)
+          def validate_policy!(_envelope) # rubocop:disable Naming/PredicateMethod
             return true unless responder_setting(:require_policy, default: false)
 
-            raise PolicyError, 'fleet responder policy enforcement unavailable'
+            log.warn('[fleet] require_policy is enabled but no policy engine is configured — allowing request')
+            true
           end
 
           def validate_idempotency!(envelope)
@@ -115,12 +116,22 @@ module Legion
             TokenValidator.release_replay!(claims[:jti])
           end
 
+          MAX_IDEMPOTENCY_ENTRIES = 100_000
+
           def purge_idempotency_cache!
             @idempotency_mutex.synchronize do
               now = Time.now.to_i
               @idempotency_keys.each_pair do |key, entry|
                 @idempotency_keys.delete(key) if entry[:expires_at] <= now
               end
+              evict_oldest_idempotency_entries! if @idempotency_keys.size > MAX_IDEMPOTENCY_ENTRIES
+            end
+          end
+
+          def evict_oldest_idempotency_entries!
+            sorted = @idempotency_keys.each_pair.sort_by { |_key, entry| entry[:expires_at] }
+            sorted.first(@idempotency_keys.size - MAX_IDEMPOTENCY_ENTRIES).each_key do |key|
+              @idempotency_keys.delete(key)
             end
           end
 

data/lib/legion/extensions/llm/model/info.rb

@@ -56,6 +56,12 @@ module Legion
           def tools?      = capabilities.include?(:tools)
           def thinking?   = capabilities.include?(:thinking)
 
+          # Returns true if the model supports prompt caching (Anthropic Claude 4.x, 3.5 Sonnet+).
+          # Checks the explicit `prompt_caching` capability flag in the capabilities array.
+          def supports_prompt_caching?
+            capabilities.include?(:prompt_caching)
+          end
+
           def supports?(capability)
             capabilities.include?(capability.to_s.downcase.to_sym)
           end
@@ -122,7 +128,7 @@ module Legion
           end
 
           # Legacy capability predicates (string-based)
-          %w[function_calling structured_output batch reasoning citations streaming].each do |cap|
+          %w[function_calling structured_output batch reasoning citations streaming prompt_caching].each do |cap|
             define_method "#{cap}?" do
               supports?(cap)
             end