riffer 0.50.0 → 0.51.0

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 1700c8028d7de9b0e109fc259f904c8a8ff1e92964e23c90cd00574cd43a0bc2
-  data.tar.gz: 1245b96f6a1490aa5cedd3d5e60ff062a8d70cc7330bd7ec01d0f027e05d65fa
+  metadata.gz: c1379919ce4dcd42bd21286bb3447f32967cd94fff23cf5a3ed2e422b40a6cfc
+  data.tar.gz: 8d7860fa41acf5e89e82bd25f8f57477da356cb6948a4438d9ca0a7c71a6d959
 SHA512:
-  metadata.gz: dc40857b3085259180f014442cd8d2ec4c6f6f85d2c577cacbff6332176eb39e948de9e20b22732730e52ea4151016fcf23c3387de181f8b8ff27537df8cc21f
-  data.tar.gz: '0152801a535973a01216aac4ff1a5b07554b2e8bde282a9dd3fc2c09408b0ddb7890953e7d60c21102d0889896190adf88d6822ef0ca0ff4a6564b688f7b4dd2'
+  metadata.gz: 882930df0b0f87cc9e2dbe3b792969db10b811a1c33d92ea5797d53564421177032c5ad75362247cbd5e012784fea7bf25c90d5de3a3ce54cc472af2c210e268
+  data.tar.gz: 9b23fe7c44c6d3fbbbe5ef165f2dc07c1a55ad2050dfd7b188cb674b9b3fe6ee45441583e8cc544092e3820c1c2a30f1f9809ce3ddd874d038180f94184990fd

data/.release-please-manifest.json

@@ -1,3 +1,3 @@
 {
-  ".": "0.50.0"
+  ".": "0.51.0"
 }

data/CHANGELOG.md

@@ -5,6 +5,25 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.51.0](https://github.com/janeapp/riffer/compare/riffer/v0.50.1...riffer/v0.51.0) (2026-10-07)
+
+
+### Features
+
+* **structured_output:** recover objects wrapped in code fences or prose ([#473](https://github.com/janeapp/riffer/issues/473)) ([59a6d40](https://github.com/janeapp/riffer/commit/59a6d40b08b9f18fcb4ebac24598925f94a9acea))
+
+
+### Bug Fixes
+
+* **agent:** apply the max_steps cap per run, not per conversation ([#478](https://github.com/janeapp/riffer/issues/478)) ([4948e6b](https://github.com/janeapp/riffer/commit/4948e6b9769df084bed8c8a29862067560f800bc))
+
+## [0.50.1](https://github.com/janeapp/riffer/compare/riffer/v0.50.0...riffer/v0.50.1) (2026-10-02)
+
+
+### Bug Fixes
+
+* **providers:** replay reasoning parts with no text as empty strings ([#470](https://github.com/janeapp/riffer/issues/470)) ([a76d355](https://github.com/janeapp/riffer/commit/a76d3550d1e030b2d77118fb33e5d52271f00393))
+
 ## [0.50.0](https://github.com/janeapp/riffer/compare/riffer/v0.49.0...riffer/v0.50.0) (2026-09-28)
 
 

data/docs/AGENTS.md

@@ -168,7 +168,9 @@ end
 
 The LLM response is automatically parsed and validated against the schema. Access the result via `response.structured_output`.
 
-When the response is not valid JSON or does not satisfy the schema, `response.structured_output` is `nil`, `response.outcome.reason` is `:invalid_structured_output`, and `response.outcome.detail` carries the parse or validation message:
+If the response is not valid JSON on its own, Riffer recovers a JSON object wrapped in a code fence (```` ```json ````), surrounded by prose (`Here is the data: {...}`), or wrapped in formatting (`**{...}**`). It parses the span from the first `{` to the last `}`, so a reply whose surrounding text contains other braces, or that holds more than one object, is still rejected; `response.content` keeps the raw text.
+
+When the response contains no well-formed JSON object or does not satisfy the schema, `response.structured_output` is `nil`, `response.outcome.reason` is `:invalid_structured_output`, and `response.outcome.detail` carries the parse or validation message:
 
 ```ruby
 response = SentimentAgent.generate('Analyze: "I love this!"')

data/docs/AGENT_LIFECYCLE.md

@@ -351,7 +351,7 @@ end
 
 ### response.structured_output
 
-When structured output is configured, the LLM response is parsed as JSON and validated against the schema. The validated result is available as `response.structured_output`:
+When structured output is configured, the LLM response is parsed as JSON and validated against the schema. A JSON object wrapped in a code fence, prose, or formatting is recovered (see [Agents — structured_output](AGENTS.md#structured_output)). The validated result is available as `response.structured_output`:
 
 ```ruby
 response = SentimentAgent.generate('Analyze: "I love this!"')

data/docs/AGENT_LOOP.md

@@ -72,12 +72,12 @@ response = agent.generate('Approved, continue')  # continues where it left off
 
 ### Max Steps Limit
 
-The `max_steps` class method caps the number of LLM call steps in the tool-use loop. When the step count reaches the limit, the loop interrupts automatically with reason `:max_steps`.
+The `max_steps` class method caps the number of LLM call steps in the tool-use loop of a single run. Steps are counted from the last user message in the session, so each new user message starts a fresh budget — prior turns' assistant messages in the history do not count against it. When the step count reaches the limit, the loop interrupts automatically with reason `:max_steps`.
 
 - **When to use:** Safety net to prevent runaway tool-use loops — useful when agents have access to many tools or operate autonomously.
 - **Response:** `response.outcome.reason` is `:max_steps`.
 - **Streaming:** Yields an `Interrupt` event with `reason: :max_steps`.
-- **Resumable:** Yes. Call `generate('Continue')` or `stream('Continue')` on the same agent instance to resume. For cross-process resume, pass persisted messages as an array to a new agent. Pending tool calls are automatically executed before the LLM loop resumes.
+- **Resumable:** Yes. Call `generate('Continue')` or `stream('Continue')` on the same agent instance to resume with a fresh budget, or call `generate` / `stream` with no prompt to resume against the remaining budget. For cross-process resume, pass persisted messages as an array to a new agent; the same rule applies. Pending tool calls are automatically executed before the LLM loop resumes.
 
 ```ruby
 class MyAgent < Riffer::Agent

data/docs/MESSAGES.md

@@ -164,7 +164,7 @@ Each part carries:
 | `data`      | `String?` | The opaque payload, for `:encrypted` parts                                                                      |
 | `signature` | `String?` | The provider's signature over the part, when it issues one                                                      |
 | `id`        | `String?` | The provider's identifier for the part, when it issues one                                                      |
-| `format`    | `String?` | The wire format, owned by the adapter that produced the part (e.g. `"anthropic-claude-v1"`)                     |
+| `format`    | `String?` | The wire format, owned by the adapter that produced the part (e.g. `"anthropic-messages-v1"`)                   |
 
 A `type` outside the three values raises `Riffer::ArgumentError`. `format` is a free string riffer never validates — it exists so an adapter can tell its own parts apart from another adapter's.
 

data/lib/riffer/agent/run.rb

@@ -68,7 +68,7 @@ module Riffer::Agent::Run
     end
 
     begin
-      step = agent.session.steps
+      step = agent.session.steps_since_last_user_message
 
       reason = catch(:riffer_interrupt) do
         execute_pending_tool_calls(agent, tags)

data/lib/riffer/agent/session.rb

@@ -143,6 +143,13 @@ class Riffer::Agent::Session
     @messages.count { |m| m.is_a?(Riffer::Messages::Assistant) }
   end
 
+  #--
+  #: () -> Integer
+  def steps_since_last_user_message
+    since_last_user = @messages.reverse_each.take_while { |m| !m.is_a?(Riffer::Messages::User) }
+    since_last_user.count { |m| m.is_a?(Riffer::Messages::Assistant) }
+  end
+
   #--
   #: () -> Riffer::Messages::Assistant?
   def final_assistant_message

data/lib/riffer/agent/structured_output/parser.rb

@@ -0,0 +1,35 @@
+# frozen_string_literal: true
+# rbs_inline: enabled
+
+require "json"
+
+module Riffer::Agent::StructuredOutput::Parser
+  extend self
+
+  #--
+  #: (String) -> Hash[Symbol, untyped]?
+  def parse(content)
+    parse_object(content) || outermost_object(content)
+  end
+
+  private
+
+  #--
+  #: (String) -> Hash[Symbol, untyped]?
+  def outermost_object(content)
+    start = content.index("{")
+    stop = content.rindex("}")
+    return unless start && stop
+
+    parse_object(content[start..stop].to_s)
+  end
+
+  #--
+  #: (String) -> Hash[Symbol, untyped]?
+  def parse_object(json)
+    parsed = JSON.parse(json, symbolize_names: true)
+    parsed if parsed.is_a?(Hash)
+  rescue JSON::ParserError
+    nil
+  end
+end

data/lib/riffer/agent/structured_output.rb

@@ -1,8 +1,6 @@
 # frozen_string_literal: true
 # rbs_inline: enabled
 
-require "json"
-
 class Riffer::Agent::StructuredOutput
   attr_reader :params #: Riffer::Params # @dynamic params
 
@@ -21,11 +19,11 @@ class Riffer::Agent::StructuredOutput
   #--
   #: (String) -> Riffer::Agent::StructuredOutput::Result
   def parse_and_validate(json_string)
-    parsed = JSON.parse(json_string, symbolize_names: true)
+    parsed = Riffer::Agent::StructuredOutput::Parser.parse(json_string)
+    return Result.new(error: "JSON parse error: no JSON object found in the response") if parsed.nil?
+
     validated = @params.validate(parsed)
     Result.new(object: validated)
-  rescue JSON::ParserError => e
-    Result.new(error: "JSON parse error: #{e.message}")
   rescue Riffer::ValidationError => e
     Result.new(error: "Validation error: #{e.message}")
   end

data/lib/riffer/messages/assistant/reasoning_part.rb

@@ -12,7 +12,7 @@ class Riffer::Messages::Assistant::ReasoningPart
   attr_reader :signature #: String? # @dynamic signature
   attr_reader :id #: String? # @dynamic id
 
-  # Owned by the producing adapter (e.g. <tt>"anthropic-claude-v1"</tt>); adapters replay only the
+  # Owned by the producing adapter (e.g. <tt>"anthropic-messages-v1"</tt>); adapters replay only the
   # formats they recognize, so it is never validated here.
   attr_reader :format #: String? # @dynamic format
 

data/lib/riffer/providers/amazon_bedrock.rb

@@ -480,7 +480,7 @@ class Riffer::Providers::AmazonBedrock < Riffer::Providers::Base
     data = part.data
     return { reasoning_content: { redacted_content: Base64.strict_decode64(data) } } if data
 
-    { reasoning_content: { reasoning_text: { text: part.text, signature: part.signature }.compact } }
+    { reasoning_content: { reasoning_text: { text: part.text.to_s, signature: part.signature }.compact } }
   end
 
   #--

data/lib/riffer/providers/anthropic.rb

@@ -440,7 +440,7 @@ class Riffer::Providers::Anthropic < Riffer::Providers::Base
   def convert_reasoning_part_to_anthropic_format(part)
     return { type: "redacted_thinking", data: part.data } if part.type == :encrypted
 
-    { type: "thinking", thinking: part.text, signature: part.signature }
+    { type: "thinking", thinking: part.text.to_s, signature: part.signature }
   end
 
   #--

data/lib/riffer/providers/base.rb

@@ -360,9 +360,7 @@ class Riffer::Providers::Base
   #--
   #: (String) -> Hash[Symbol, untyped]?
   def parse_structured_output(content)
-    JSON.parse(content, symbolize_names: true)
-  rescue JSON::ParserError
-    nil
+    Riffer::Agent::StructuredOutput::Parser.parse(content)
   end
 
   #--

data/lib/riffer/providers/open_ai.rb

@@ -446,8 +446,8 @@ class Riffer::Providers::OpenAI < Riffer::Providers::Base
   #--
   #: (Array[Riffer::Messages::Assistant::ReasoningPart]) -> Hash[Symbol, untyped]
   def convert_reasoning_item_to_openai_format(parts)
-    summary = parts.filter_map { |part| { type: "summary_text", text: part.text } if part.type == :summary }
-    content = parts.filter_map { |part| { type: "reasoning_text", text: part.text } if part.type == :text }
+    summary = parts.filter_map { |part| { type: "summary_text", text: part.text.to_s } if part.type == :summary }
+    content = parts.filter_map { |part| { type: "reasoning_text", text: part.text.to_s } if part.type == :text }
 
     {
       type: "reasoning",

data/lib/riffer/version.rb

@@ -2,5 +2,5 @@
 # rbs_inline: enabled
 
 module Riffer
-  VERSION = "0.50.0" #: String
+  VERSION = "0.51.0" #: String
 end

data/rbs_collection.lock.yaml

@@ -10,7 +10,7 @@ gems:
     remote: https://github.com/ruby/gem_rbs_collection.git
     repo_dir: gems
 - name: anthropic
-  version: 1.71.0
+  version: 1.76.0
   source:
     type: rubygems
 - name: ast
@@ -170,7 +170,7 @@ gems:
   source:
     type: stdlib
 - name: openai
-  version: 0.91.0
+  version: 0.97.0
   source:
     type: rubygems
 - name: openssl

data/sig/generated/riffer/agent/session.rbs

@@ -57,6 +57,10 @@ class Riffer::Agent::Session
   # : () -> Integer
   def steps: () -> Integer
 
+  # --
+  # : () -> Integer
+  def steps_since_last_user_message: () -> Integer
+
   # --
   # : () -> Riffer::Messages::Assistant?
   def final_assistant_message: () -> Riffer::Messages::Assistant?

data/sig/generated/riffer/agent/structured_output/parser.rbs

@@ -0,0 +1,17 @@
+# Generated from lib/riffer/agent/structured_output/parser.rb with RBS::Inline
+
+module Riffer::Agent::StructuredOutput::Parser
+  # --
+  # : (String) -> Hash[Symbol, untyped]?
+  def parse: (String) -> Hash[Symbol, untyped]?
+
+  private
+
+  # --
+  # : (String) -> Hash[Symbol, untyped]?
+  def outermost_object: (String) -> Hash[Symbol, untyped]?
+
+  # --
+  # : (String) -> Hash[Symbol, untyped]?
+  def parse_object: (String) -> Hash[Symbol, untyped]?
+end

data/sig/generated/riffer/messages/assistant/reasoning_part.rbs

@@ -15,7 +15,7 @@ class Riffer::Messages::Assistant::ReasoningPart
 
   attr_reader id: String?
 
-  # Owned by the producing adapter (e.g. <tt>"anthropic-claude-v1"</tt>); adapters replay only the
+  # Owned by the producing adapter (e.g. <tt>"anthropic-messages-v1"</tt>); adapters replay only the
   # formats they recognize, so it is never validated here.
   attr_reader format: String?
 

data/sig/manual/riffer/agent/structured_output/parser.rbs

@@ -0,0 +1,5 @@
+# `Riffer::Agent::StructuredOutput::Parser` uses `extend self`; rbs-inline doesn't emit that,
+# so re-extend here to expose its instance methods as singleton methods.
+module Riffer::Agent::StructuredOutput::Parser
+  extend ::Riffer::Agent::StructuredOutput::Parser
+end

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: riffer
 version: !ruby/object:Gem::Version
-  version: 0.50.0
+  version: 0.51.0
 platform: ruby
 authors:
 - Jake Bottrall
@@ -117,6 +117,7 @@ files:
 - lib/riffer/agent/session.rb
 - lib/riffer/agent/session/repair.rb
 - lib/riffer/agent/structured_output.rb
+- lib/riffer/agent/structured_output/parser.rb
 - lib/riffer/agent/structured_output/result.rb
 - lib/riffer/config.rb
 - lib/riffer/config/amazon_bedrock.rb
@@ -265,6 +266,7 @@ files:
 - sig/generated/riffer/agent/session.rbs
 - sig/generated/riffer/agent/session/repair.rbs
 - sig/generated/riffer/agent/structured_output.rbs
+- sig/generated/riffer/agent/structured_output/parser.rbs
 - sig/generated/riffer/agent/structured_output/result.rbs
 - sig/generated/riffer/config.rbs
 - sig/generated/riffer/config/amazon_bedrock.rbs
@@ -390,6 +392,7 @@ files:
 - sig/manual/riffer/agent/run.rbs
 - sig/manual/riffer/agent/serializer.rbs
 - sig/manual/riffer/agent/session/repair.rbs
+- sig/manual/riffer/agent/structured_output/parser.rbs
 - sig/manual/riffer/evals/evaluator_runner.rbs
 - sig/manual/riffer/helpers/boolean.rbs
 - sig/manual/riffer/helpers/call_or_value.rbs