Skip to content

fix(google): handle more thoughtSignature vagaries during streaming #7204

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
DOsinga merged 1 commit into from
Feb 17, 2026
Merged

fix(google): handle more thoughtSignature vagaries during streaming #7204

Changes from all commits
Commits
Show all changes
1 commit
Select commit
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
164 changes: 55 additions & 109 deletions crates/goose/src/providers/formats/google.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -227,38 +227,6 @@ enum SignedTextHandling {
SignedTextAsRegularText,
}

pub fn process_response_part(
part: &Value,
last_signature: &mut Option<String>,
) -> Option<MessageContent> {
let has_signature = part.get(THOUGHT_SIGNATURE_KEY).is_some();
let handling = if has_signature {
SignedTextHandling::SignedTextAsThinking
} else {
SignedTextHandling::SignedTextAsRegularText
};
process_response_part_impl(part, last_signature, handling)
}

/// Gemini 2.x includes thoughtSignature on first chunk as metadata, not actual thinking.
fn process_response_part_for_model(
part: &Value,
last_signature: &mut Option<String>,
model_version: Option<&str>,
) -> Option<MessageContent> {
let is_gemini_2 = model_version
.map(|m| m.starts_with("gemini-2"))
.unwrap_or(false);

let has_signature = part.get(THOUGHT_SIGNATURE_KEY).is_some();
let handling = if has_signature && !is_gemini_2 {
SignedTextHandling::SignedTextAsThinking
} else {
SignedTextHandling::SignedTextAsRegularText
};
process_response_part_impl(part, last_signature, handling)
}

fn process_response_part_non_streaming(
part: &Value,
last_signature: &mut Option<String>,
Expand Down Expand Up @@ -488,8 +456,6 @@ where
}
}

let model_version = chunk.get("modelVersion").and_then(|v| v.as_str());

let parts = chunk
.get("candidates")
.and_then(|v| v.as_array())
Expand All @@ -500,7 +466,9 @@ where

if let Some(parts) = parts {
for part in parts {
if let Some(content) = process_response_part_for_model(part, &mut last_signature, model_version) {
// Always emit text as regular text during streaming — we can't
// know yet whether function calls will follow.
if let Some(content) = process_response_part_impl(part, &mut last_signature, SignedTextHandling::SignedTextAsRegularText) {
let message = Message::new(
Role::Assistant,
chrono::Utc::now().timestamp(),
Expand Down Expand Up @@ -1192,90 +1160,68 @@ mod tests {
async fn test_streaming_with_thought_signature() {
use futures::StreamExt;

let gemini3_stream = concat!(
r#"data: {"candidates": [{"content": {"role": "model", "#,
r#""parts": [{"text": "Begin", "thoughtSignature": "sig123"}]}}], "#,
r#""modelVersion": "gemini-3-pro"}"#,
"\n",
r#"data: {"candidates": [{"content": {"role": "model", "#,
r#""parts": [{"text": " end"}]}}], "modelVersion": "gemini-3-pro"}"#
);
let lines: Vec<Result<String, anyhow::Error>> =
gemini3_stream.lines().map(|l| Ok(l.to_string())).collect();
let stream = Box::pin(futures::stream::iter(lines));
let mut message_stream = std::pin::pin!(response_to_streaming_message(stream));

let mut text_parts = Vec::new();
let mut thinking_parts = Vec::new();

while let Some(result) = message_stream.next().await {
let (message, _usage) = result.unwrap();
if let Some(msg) = message {
match msg.content.first() {
Some(MessageContent::Text(text)) => text_parts.push(text.text.clone()),
Some(MessageContent::Thinking(t)) => thinking_parts.push(t.thinking.clone()),
_ => {}
async fn collect_streaming_text(raw: &str) -> (String, usize) {
let lines: Vec<Result<String, anyhow::Error>> =
raw.lines().map(|l| Ok(l.to_string())).collect();
let stream = Box::pin(futures::stream::iter(lines));
let mut msg_stream = std::pin::pin!(response_to_streaming_message(stream));
let mut text = String::new();
let mut thinking = 0usize;
while let Some(Ok((message, _))) = msg_stream.next().await {
if let Some(msg) = message {
for c in &msg.content {
match c {
MessageContent::Text(t) => text.push_str(&t.text),
MessageContent::Thinking(_) => thinking += 1,
_ => {}
}
}
}
}
(text, thinking)
}

assert_eq!(thinking_parts, vec!["Begin"]);
assert_eq!(text_parts, vec![" end"]);

let gemini25_stream = concat!(
// First chunk signed
let (text, thinking) = collect_streaming_text(concat!(
r#"data: {"candidates": [{"content": {"role": "model", "#,
r#""parts": [{"text": "Begin", "thoughtSignature": "sig123"}]}}], "#,
r#""modelVersion": "gemini-2.5-pro"}"#,
r#""parts": [{"text": "Hello", "thoughtSignature": "sig1"}]}}], "#,
r#""modelVersion": "gemini-3-flash-preview"}"#,
"\n",
r#"data: {"candidates": [{"content": {"role": "model", "#,
r#""parts": [{"text": " end"}]}}], "modelVersion": "gemini-2.5-pro"}"#
);
let lines: Vec<Result<String, anyhow::Error>> =
gemini25_stream.lines().map(|l| Ok(l.to_string())).collect();
let stream = Box::pin(futures::stream::iter(lines));
let mut message_stream = std::pin::pin!(response_to_streaming_message(stream));

let mut text_parts = Vec::new();

while let Some(result) = message_stream.next().await {
let (message, _usage) = result.unwrap();
if let Some(msg) = message {
if let Some(MessageContent::Text(text)) = msg.content.first() {
text_parts.push(text.text.clone());
}
}
}

assert_eq!(text_parts, vec!["Begin", " end"]);

let unknown_stream = concat!(
r#""parts": [{"text": " world"}]}}], "modelVersion": "gemini-3-flash-preview"}"#
))
.await;
assert_eq!(thinking, 0);
assert_eq!(text, "Hello world");

// Last chunk signed (the reported truncation bug)
let (text, thinking) = collect_streaming_text(concat!(
r#"data: {"candidates": [{"content": {"role": "model", "#,
r#""parts": [{"text": "Begin", "thoughtSignature": "sig123"}]}}]}"#,
r#""parts": [{"text": "SECURITY.md: Project"}]}}], "#,
r#""modelVersion": "gemini-3-flash-preview"}"#,
"\n",
r#"data: {"candidates": [{"content": {"role": "model", "#,
r#""parts": [{"text": " end"}]}}]}"#
);
let lines: Vec<Result<String, anyhow::Error>> =
unknown_stream.lines().map(|l| Ok(l.to_string())).collect();
let stream = Box::pin(futures::stream::iter(lines));
let mut message_stream = std::pin::pin!(response_to_streaming_message(stream));

let mut text_parts = Vec::new();
let mut thinking_parts = Vec::new();

while let Some(result) = message_stream.next().await {
let (message, _usage) = result.unwrap();
if let Some(msg) = message {
match msg.content.first() {
Some(MessageContent::Text(text)) => text_parts.push(text.text.clone()),
Some(MessageContent::Thinking(t)) => thinking_parts.push(t.thinking.clone()),
_ => {}
}
}
}

assert_eq!(thinking_parts, vec!["Begin"]);
assert_eq!(text_parts, vec![" end"]);
r#""parts": [{"text": " policies.\n\nRead it?", "thoughtSignature": "sig2"}]}}], "#,
r#""modelVersion": "gemini-3-flash-preview"}"#
))
.await;
assert_eq!(thinking, 0);
assert_eq!(text, "SECURITY.md: Project policies.\n\nRead it?");

// Intermediate chunk signed
let (text, thinking) = collect_streaming_text(concat!(
r#"data: {"candidates": [{"content": {"role": "model", "#,
r#""parts": [{"text": "one "}]}}], "modelVersion": "gemini-3-flash-preview"}"#,
"\n",
r#"data: {"candidates": [{"content": {"role": "model", "#,
r#""parts": [{"text": "two ", "thoughtSignature": "sig3"}]}}], "modelVersion": "gemini-3-flash-preview"}"#,
"\n",
r#"data: {"candidates": [{"content": {"role": "model", "#,
r#""parts": [{"text": "three"}]}}], "modelVersion": "gemini-3-flash-preview"}"#
))
.await;
assert_eq!(thinking, 0);
assert_eq!(text, "one two three");
}

#[tokio::test]
Expand Down