legion-llm 0.8.22 → 0.8.24

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 06c8e0373f627e588f41b1c2d8fda18fd75bbc0e673ab270ec0d08bfe27695ec
-  data.tar.gz: 15c2e48761d2e797e144178db61363344663c9860918ddd001088532eaad84ee
+  metadata.gz: 1c7a3d39cf4e2e31494ee7e354680e57ae1c6c1feabe2f39e8707a06701c6a15
+  data.tar.gz: c8733d7f96801aa19c35458f4e527574815909dd87e89ed80d589bf565a8467c
 SHA512:
-  metadata.gz: e4420346b02d8ec03fb5b80d930f678256245b4b80bd75076e9be15e9abf82d3f569af16f0278cbc78ec0bf8deb6058484045beb27fee8db265f8abf37a67d87
-  data.tar.gz: e1477099d25547be1bd2f3556a19d1d38a6ce7c77eb3b856c03c0605d1d2510eaa173828e63ed7e1d2763dc2d4f036e754bc93c274b5a98b3437de72bc72aeaf
+  metadata.gz: 37049fdb4a5dc838fecc0d3b6c57e48bbcb72d490b8e8460e04cdd7a19728d82e8c2c3f48ab0bfc059bc2245d56f40f279d79b082e31c7f4c85fa57d86270e42
+  data.tar.gz: 06ae35daf7458e38b4990c2a230a5f72bffd65bc99bc3dd2bbb16be4ff60aa3665c3b807c3b19637f407fc1e55b6f88148c433d39474411bfb222fe9e07d412c

data/CHANGELOG.md

@@ -1,5 +1,17 @@
 # Legion LLM Changelog
 
+## [0.8.24] - 2026-04-23
+
+### Fixed
+- All AMQP transport messages (audit, metering, tool, escalation) now include identity headers (`x-legion-identity`, `x-legion-credential`, `x-legion-hostname`) extracted from the `caller` field. Previously only prompt audit events carried identity in the body — tool audit and metering messages had no identity at all.
+- Embedding metering events now include `caller` context.
+- Non-pipeline `chat_single` metering events now include `caller` context from kwargs.
+
+## [0.8.23] - 2026-04-23
+
+### Fixed
+- `Call::StructuredOutput` prompt-fallback path passed `messages:` (plural) to `chat_single` which only accepts `message:` (singular), leaking the unknown kwarg into `RubyLLM::Chat.new`. Visible as repeated "unknown keyword: :messages" warnings during dream cycle contradiction detection. Flattened instruction + messages into a single string via `extract_user_content`.
+
 ## [0.8.22] - 2026-04-22
 
 ### Fixed

data/lib/legion/llm/call/embeddings.rb

@@ -462,9 +462,16 @@ module Legion
           end
 
           def emit_embedding_metering(provider:, model:, tokens:)
+            caller = begin
+              Legion::LLM.settings[:caller]
+            rescue StandardError => e
+              handle_exception(e, level: :debug, operation: 'llm.embeddings.metering.caller')
+              nil
+            end
             Legion::LLM::Metering.emit(
               provider: provider, model_id: model, request_type: 'embed',
-              tier: 'cloud', input_tokens: tokens.to_i, output_tokens: 0, total_tokens: tokens.to_i
+              tier: 'cloud', input_tokens: tokens.to_i, output_tokens: 0, total_tokens: tokens.to_i,
+              caller: caller
             )
           rescue StandardError => e
             handle_exception(e, level: :warn, operation: 'llm.embeddings.metering')

data/lib/legion/llm/call/structured_output.rb

@@ -36,10 +36,10 @@ module Legion
               instruction = "You MUST respond with valid JSON matching this schema:\n" \
                             "```json\n#{Legion::JSON.dump(schema)}\n```\n" \
                             'Respond with ONLY the JSON object, no other text.'
-              augmented = [{ role: 'system', content: instruction }] + Array(messages)
+              user_content = extract_user_content(messages, instruction)
               Legion::LLM::Inference.send(:chat_single,
                                           model: model, provider: provider, intent: nil, tier: nil,
-                                          messages: augmented, **opts.except(:attempt))
+                                          message: user_content, **opts.except(:attempt))
             end
           end
 
@@ -55,10 +55,10 @@ module Legion
 
           def retry_with_instruction(messages, schema, model, provider: nil, **opts)
             instruction = "Your previous response was not valid JSON. Respond with ONLY a valid JSON object matching this schema:\n#{Legion::JSON.dump(schema)}"
-            augmented = Array(messages) + [{ role: 'user', content: instruction }]
+            user_content = extract_user_content(messages, instruction)
             result = Legion::LLM::Inference.send(:chat_single,
                                                  model: model, provider: provider, intent: nil, tier: nil,
-                                                 messages: augmented, **opts.except(:attempt))
+                                                 message: user_content, **opts.except(:attempt))
 
             parsed = Legion::JSON.load(result[:content])
             { data: parsed, raw: result[:content], model: result[:model], valid: true, retried: true }
@@ -67,6 +67,15 @@ module Legion
             { data: nil, error: e.message, valid: false }
           end
 
+          def extract_user_content(messages, instruction)
+            parts = [instruction]
+            Array(messages).each do |msg|
+              content = msg[:content] || msg['content']
+              parts << content.to_s unless content.to_s.empty?
+            end
+            parts.join("\n\n")
+          end
+
           def supports_response_format?(model)
             SCHEMA_CAPABLE_MODELS.any? { |m| model.to_s.include?(m) }
           end

data/lib/legion/llm/inference.rb

@@ -514,7 +514,7 @@ module Legion
         log.debug '[llm][inference] chat_single asking session'
         response = block ? session.ask(message, &block) : session.ask(message)
         log.debug "[llm][inference] chat_single response_class=#{response.class} response_nil=#{response.nil?}"
-        emit_non_pipeline_metering(response, model: opts[:model], provider: opts[:provider])
+        emit_non_pipeline_metering(response, model: opts[:model], provider: opts[:provider], caller: kwargs[:caller])
 
         if response && !block && defined?(Quality::ShadowEval) && Quality::ShadowEval.enabled?
           msgs = session.respond_to?(:messages) ? session.messages : nil
@@ -712,14 +712,15 @@ module Legion
         esc.fetch(:quality_threshold, 50)
       end
 
-      def emit_non_pipeline_metering(response, model:, provider:)
+      def emit_non_pipeline_metering(response, model:, provider:, caller: nil)
         return unless response
 
         input  = response.respond_to?(:input_tokens)  ? response.input_tokens.to_i  : 0
         output = response.respond_to?(:output_tokens) ? response.output_tokens.to_i : 0
         Legion::LLM::Metering.emit(
           provider: provider, model_id: model, request_type: 'chat',
-          tier: 'direct', input_tokens: input, output_tokens: output, total_tokens: input + output
+          tier: 'direct', input_tokens: input, output_tokens: output, total_tokens: input + output,
+          caller: caller
         )
       rescue StandardError => e
         handle_exception(e, level: :warn, operation: 'llm.inference.non_pipeline_metering')

data/lib/legion/llm/transport/message.rb

@@ -65,6 +65,19 @@ module Legion
           h['x-legion-llm-model']          = model_val.to_s                     if model_val
           h['x-legion-llm-request-type']   = @options[:request_type].to_s       if @options[:request_type]
           h['x-legion-llm-schema-version'] = '1.0.0'
+          h.merge(identity_headers)
+        end
+
+        def identity_headers
+          caller = @options[:caller]
+          return {} unless caller.is_a?(Hash)
+
+          rb = caller[:requested_by] || caller['requested_by'] || {}
+          h = {}
+          identity = rb[:identity] || rb['identity'] || rb[:username] || rb['username']
+          h['x-legion-identity']   = identity.to_s   if identity
+          h['x-legion-credential'] = (rb[:credential] || rb['credential']).to_s if rb[:credential] || rb['credential']
+          h['x-legion-hostname']   = (rb[:hostname] || rb['hostname']).to_s     if rb[:hostname] || rb['hostname']
           h
         end
 

data/lib/legion/llm/version.rb

@@ -2,6 +2,6 @@
 
 module Legion
   module LLM
-    VERSION = '0.8.22'
+    VERSION = '0.8.24'
   end
 end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: legion-llm
 version: !ruby/object:Gem::Version
-  version: 0.8.22
+  version: 0.8.24
 platform: ruby
 authors:
 - Esity