ask-agent 0.4.5 → 0.4.6

checksums.yaml

@@ -1,7 +1,7 @@
 ---
 SHA256:
-  metadata.gz: 84cfe80b10d8495e3d0c1235f63767064b9615a740991b734652cccd0a112abe
-  data.tar.gz: 95042e3fd3530a683b81259445afc524674808278e48fe7f90a7a89bb8e7f22c
+  metadata.gz: 56dc1b3abe92b66b571ee60669c0b95b9a6b00f5ae0b83d8a2777b2d10bb53dc
+  data.tar.gz: ea70a273c33e7c04dc9e8dc0e32117e8be60bf780106f8bf36a0f83b7897a73e
 SHA512:
-  metadata.gz: af3eaf7b48c3ecb516977f4182df9ec19ae640d14c94cee2173cb53cb78d8cafc8d654a8a4b8c99065413e664e0a7ec73ec35bc63ba1c803093b3cce6bf15de9
-  data.tar.gz: 2a59313a60550a171b60e1c6a95da02d049f72f6aacabd6cc0bb86385f120bcdb8401bd82011098b6b162ae72bdb2cfda5543b1776ed02f0ee2bb0ca432ba8f9
+  metadata.gz: 43298ab2aa83f7e16c86498d6e3f15d1b01821d81794cd2023574e1553446755913982d9323e8f5ff4fc603982570a21734a9b8947ab2c8d34982a45676cd2ad
+  data.tar.gz: f95818107b8b50b612944a7e2acb6fde05ef72becc8752866106f1de495e93ddf05b88695c7dc29af1e253bf603f6da0b0cf1cbaf7d3308e62da006201d6c04b

data/CHANGELOG.md

@@ -1,3 +1,21 @@
+## [0.4.5] — 2026-07-18
+
+### Fixed
+
+- **`ToolExecutor#try_call` now respects `Ask::Result#ok?` for error detection** — Previously the method always set `is_error: false`, treating all Ask::Result returns as successful even when `ok?` was false. Tool failures returned via `Ask::Result.failure(...)` are now properly detected as errors, preventing the agent from silently ignoring failed tool executions and looping.
+
+## [0.4.4] — 2026-07-18
+
+### Added
+
+- **`ToolExecutor` detects `halted: true` from tool results and stops execution** — When a tool returns `Ask::Result.ok(metadata: { halted: true })`, the executor now detects this flag, aborts sibling tools in parallel mode, and stops sequential execution. Previously the `halted` metadata was set but never checked by the executor, causing the agent loop to continue calling tools after a tool signaled completion.
+
+## [0.4.3] — 2026-07-18
+
+### Added
+
+- **`Chat#provider_config` passes multiple credential names and path segments to `Ask::Auth.resolve`** — For compound provider slugs like `opencode_go`, the method now tries flat key names (`:opencode_go_api_key`, `:opencode_api_key`) and path segments (`[:opencode, :go, :api_key]`, `[:opencode, :api_key]`) as fallbacks. This lets `Ask::Auth.resolve` find credentials stored under various naming conventions.
+
 ## [0.4.0] — 2026-07-17
 
 ### Added

data/lib/ask/agent/session.rb

@@ -273,9 +273,15 @@ module Ask
       end
 
       def resolve_tools(tools)
-        tools.map do |tool|
+        resolved = tools.map do |tool|
           tool.is_a?(Class) ? tool.new : tool
         end
+        # Always include the load_skill tool for progressive skill disclosure,
+        # unless the test framework is loaded (test mode keeps tools deterministic)
+        unless defined?(Ask::Agent::Test) && Ask::Agent::Test
+          resolved << Skills::LoadSkillTool.new(registry: @skills_registry) unless resolved.any? { |t| t.name == "load_skill" }
+        end
+        resolved
       end
 
       def build_compactor(config)

data/lib/ask/agent/skills/load_skill_tool.rb

@@ -0,0 +1,39 @@
+# frozen_string_literal: true
+
+module Ask
+  module Agent
+    module Skills
+      # Built-in tool that lets the LLM load a discovered skill's full
+      # instructions on demand. Skills are listed by name and description
+      # via progressive disclosure — the LLM decides which to load.
+      #
+      # Usage by the LLM: call load_skill with the skill name.
+      class LoadSkillTool < Ask::Tool
+        description "Load the full instructions for a skill by name. Use this when a listed skill seems relevant to the current task."
+
+        param :name, type: :string, desc: "Name of the skill to load (e.g., writing-guide)", required: true
+
+        def initialize(registry:)
+          @registry = registry
+          super()
+        end
+
+        def execute(name:)
+          skill = @registry&.[](name)
+          unless skill
+            available = @registry&.names&.join(", ") || "none"
+            return Ask::Result.failure(message: "Skill '#{name}' not found. Available skills: #{available}")
+          end
+
+          # Return the full skill content so it can be injected into conversation
+          content = "## Skill: #{skill.name}\n#{skill.description}\n\n#{skill.instructions}"
+          Ask::Result.ok(data: { name: skill.name, content: content })
+        end
+
+        def name
+          "load_skill"
+        end
+      end
+    end
+  end
+end

data/lib/ask/agent/version.rb

@@ -2,6 +2,6 @@
 
 module Ask
   module Agent
-    VERSION = "0.4.5"
+    VERSION = "0.4.6"
   end
 end

data/lib/ask/agent.rb

@@ -56,6 +56,7 @@ require_relative "agent/configuration"
 require_relative "agent/meta_agent"
 require_relative "agent/persistence/base"
 require_relative "agent/persistence/in_memory"
+require_relative "agent/skills/load_skill_tool"
 
 # Test helpers (loaded on demand)
 autoload :Test, "ask/agent/test"

metadata

@@ -1,7 +1,7 @@
 --- !ruby/object:Gem::Specification
 name: ask-agent
 version: !ruby/object:Gem::Version
-  version: 0.4.5
+  version: 0.4.6
 platform: ruby
 authors:
 - Kaka Ruto
@@ -148,6 +148,7 @@ files:
 - lib/ask/agent/persistence/in_memory.rb
 - lib/ask/agent/reflector.rb
 - lib/ask/agent/session.rb
+- lib/ask/agent/skills/load_skill_tool.rb
 - lib/ask/agent/telemetry.rb
 - lib/ask/agent/test.rb
 - lib/ask/agent/tool_abort_controller.rb