v0.11.1

Fixed

• Qwen2.5-Coder tool calls now work on llama.cpp/Ollama. Small local coder
  models emit tool calls as <tools>{…}</tools> / <tool_call>{…}</tool_call>
  text in the message content rather than as native tool_calls. The text
  recovery parser only understood bare or ```json-fenced JSON, so these calls
  were treated as prose and the model never acted (an end-to-end run against
  qwen2.5-coder-1.5b produced no file). The parser now strips the XML tool tags
  and extracts a string-aware balanced JSON object, so a content field full of
  CSS `{ }` braces no longer truncates the parse. Also recovers a bare JSON
  object embedded after a leading sentence.

What's Changed

• fix(agent): parse Qwen-Coder tool calls from small local models (0.11.1) by @geaglin in #43

Full Changelog: v0.11.0...v0.11.1

Fixed

• Qwen2.5-Coder tool calls now work on llama.cpp/Ollama. Small local coder
  models emit tool calls as <tools>{…}</tools> / <tool_call>{…}</tool_call>
  text in the message content rather than as native tool_calls. The text
  recovery parser only understood bare or ```json-fenced JSON, so these calls
  were treated as prose and the model never acted (an end-to-end run against
  qwen2.5-coder-1.5b produced no file). The parser now strips the XML tool tags
  and extracts a string-aware balanced JSON object, so a content field full of
  CSS `{ }` braces no longer truncates the parse. Also recovers a bare JSON
  object embedded after a leading sentence.

What's Changed

• fix(agent): parse Qwen-Coder tool calls from small local models (0.11.1) by @geaglin in #43

Full Changelog: v0.11.0...v0.11.1

Fixed

• Qwen2.5-Coder tool calls now work on llama.cpp/Ollama. Small local coder
  models emit tool calls as <tools>{…}</tools> / <tool_call>{…}</tool_call>
  text in the message content rather than as native tool_calls. The text
  recovery parser only understood bare or ```json-fenced JSON, so these calls
  were treated as prose and the model never acted (an end-to-end run against
  qwen2.5-coder-1.5b produced no file). The parser now strips the XML tool tags
  and extracts a string-aware balanced JSON object, so a content field full of
  CSS `{ }` braces no longer truncates the parse. Also recovers a bare JSON
  object embedded after a leading sentence.

What's Changed

• fix(agent): parse Qwen-Coder tool calls from small local models (0.11.1) by @geaglin in #43

Full Changelog: v0.11.0...v0.11.1

Fixed

• Qwen2.5-Coder tool calls now work on llama.cpp/Ollama. Small local coder
  models emit tool calls as <tools>{…}</tools> / <tool_call>{…}</tool_call>
  text in the message content rather than as native tool_calls. The text
  recovery parser only understood bare or ```json-fenced JSON, so these calls
  were treated as prose and the model never acted (an end-to-end run against
  qwen2.5-coder-1.5b produced no file). The parser now strips the XML tool tags
  and extracts a string-aware balanced JSON object, so a content field full of
  CSS `{ }` braces no longer truncates the parse. Also recovers a bare JSON
  object embedded after a leading sentence.

What's Changed

• fix(agent): parse Qwen-Coder tool calls from small local models (0.11.1) by @geaglin in #43

Full Changelog: v0.11.0...v0.11.1