fix: handle reasoning/thinking content from models

[TheArchitectit]

Fix: Handle reasoning/thinking content from models

Problem

When using reasoning-capable models (e.g., Claude with extended thinking, Grok 3, OpenAI o3/o4), the application fails with:

❌ Request failed
[error-kind: unknown]
error: assistant stream produced no content

This occurs when the model returns only thinking/reasoning blocks without regular text content.

Root Cause (Two Places!)

1. tools crate (tools/src/lib.rs)

The SSE stream parser was explicitly ignoring thinking content blocks:

OutputContentBlock::Thinking { .. } | OutputContentBlock::RedactedThinking { .. } => {}
ContentBlockDelta::ThinkingDelta { .. } | ContentBlockDelta::SignatureDelta { .. } => {}

2. CLI crate (rusty-claude-cli/src/main.rs)

CRITICAL: The CLI uses its own AnthropicRuntimeClient with separate stream processing that was ALSO ignoring thinking content - only rendering it visually without emitting events:

ContentBlockDelta::ThinkingDelta { .. } => {
    render_thinking_block_summary(out, None, false)?;  // Just visual, no event!
}

When a model returned only thinking content, zero AssistantEvent content events were produced, causing build_assistant_message to fail.

Solution

1. Runtime (runtime/src/conversation.rs)

• Added AssistantEvent::ThinkingDelta { thinking, signature } variant
• Added flush_thinking_block() helper to convert thinking to <thinking> tags
• Updated "no content" check to accept thinking as valid content

2. Tools crate (tools/src/lib.rs)

• push_output_block() now emits ThinkingDelta for thinking blocks
• ContentBlockDelta handler processes ThinkingDelta and SignatureDelta
• Synthetic MessageStop check includes thinking as valid content

3. CLI crate (rusty-claude-cli/src/main.rs)

CRITICAL FIX:

• ContentBlockDelta::ThinkingDelta now emits AssistantEvent::ThinkingDelta
• ContentBlockDelta::SignatureDelta now emits AssistantEvent::ThinkingDelta
• push_output_block() now emits ThinkingDelta for OutputContentBlock::Thinking
• Synthetic MessageStop check includes ThinkingDelta as valid content

Changes Checklist

File	Change	Why
runtime/src/conversation.rs	Added ThinkingDelta variant to AssistantEvent	Allow thinking content to flow through the runtime
runtime/src/conversation.rs	Added flush_thinking_block() helper	Convert accumulated thinking to displayable text blocks
runtime/src/conversation.rs	Updated build_assistant_message()	Accept thinking as valid content; prevent false "no content" errors
runtime/src/conversation.rs	Added tests for thinking content	Verify fix works for thinking-only and thinking+signature cases
tools/src/lib.rs	Updated push_output_block()	Emit thinking events instead of ignoring
tools/src/lib.rs	Updated ContentBlockDelta handler	Process thinking deltas and signatures
tools/src/lib.rs	Updated synthetic stop check	Treat thinking as valid content for stream completion
rusty-claude-cli/src/main.rs	CRITICAL: Emit ThinkingDelta from ContentBlockDelta::ThinkingDelta	CLI was only rendering, not emitting events
rusty-claude-cli/src/main.rs	CRITICAL: Emit ThinkingDelta from ContentBlockDelta::SignatureDelta	Capture signature events
rusty-claude-cli/src/main.rs	CRITICAL: Emit ThinkingDelta from push_output_block()	Handle non-streaming thinking blocks
rusty-claude-cli/src/main.rs	Updated synthetic stop check	Treat thinking as valid content
commands/src/lib.rs	Fixed Team command position	Alphabetical ordering for consistency
runtime/src/sandbox.rs	Added GitHub CLI env passthrough	Enable gh CLI usage within sandbox

Testing

• Added build_assistant_message_accepts_thinking_content test
• Added build_assistant_message_accepts_thinking_with_signature test
• All 23 conversation tests pass
• All 562 runtime tests pass

Impact

• Enables use of reasoning models that return thinking content
• Backward compatible: regular text/tool content flows unchanged
• Redacted thinking is intentionally skipped (no useful content to display)
• No breaking changes to public APIs