[Bug] Streaming ignores `delta.reasoning_content`, so reasoning models stream nothing / yield unparseable output

[404-Page-Found]

Description

OpenAI-compatible reasoning models (e.g. OpenAI o-series, DeepSeek-R1 via DeepSeek) stream their chain-of-thought in choices[].delta.reasoning_content and may emit content only at the end — or never. parseOpenAiSseLine only reads delta.content (sse.ts:91), so with suggest --stream:

• The reasoning phase streams no visible text (terminal appears frozen for potentially minutes).
• For providers/models that only emit reasoning_content, the accumulated string is empty and parseSuggestions() returns nothing → Could not parse any suggestions from LLM response.

Location

src/providers/sse.ts lines 88–100 (parseOpenAiSseLine)

Code

const content = parsed.choices?.[0]?.delta?.content;
if (content) result.text = content;
// delta.reasoning_content is never examined

Suggested fix

Fall back to delta.reasoning_content (optionally with a different chunk kind, or surfaced as text) when content is absent, so the user sees progressive output and parsing has text to work with. Add a unit test using a reasoning_content-only delta stream.

Impact

Broken/blank streaming UX for reasoning models, and hard failures for models that never send content deltas.