Skip to content

fix: capture text from intermediate tool_use iterations #989

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

Open
letzdoo-js wants to merge 1 commit into
base: main
Choose a base branch
from
Open

fix: capture text from intermediate tool_use iterations #989

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
93 changes: 69 additions & 24 deletions crates/openfang-runtime/src/agent_loop.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -332,6 +332,10 @@ pub async fn run_agent_loop(

let mut total_usage = TokenUsage::default();
let final_response;
// Accumulate text from intermediate iterations (tool_use turns may include text
// alongside tool calls — this text would otherwise be lost when the final
// EndTurn iteration has empty text).
let mut accumulated_text = String::new();

// Safety valve: trim excessively long message histories to prevent context overflow.
// The full compaction system handles sophisticated summarization, but this prevents
Expand Down Expand Up @@ -524,20 +528,30 @@ pub async fn run_agent_loop(
}
}

// Guard against empty response — covers both iteration 0 and post-tool cycles
// Guard against empty response — covers both iteration 0 and post-tool cycles.
// Use accumulated_text from intermediate tool_use iterations as fallback.
let text = if text.trim().is_empty() {
warn!(
agent = %manifest.name,
iteration,
input_tokens = total_usage.input_tokens,
output_tokens = total_usage.output_tokens,
messages_count = messages.len(),
"Empty response from LLM — guard activated"
);
if any_tools_executed {
"[Task completed — the agent executed tools but did not produce a text summary.]".to_string()
if !accumulated_text.is_empty() {
debug!(
agent = %manifest.name,
accumulated_len = accumulated_text.len(),
"Using accumulated text from intermediate tool_use iterations"
);
accumulated_text.clone()
} else {
"[The model returned an empty response. This usually means the model is overloaded, the context is too large, or the API key lacks credits. Try again or check /status.]".to_string()
warn!(
agent = %manifest.name,
iteration,
input_tokens = total_usage.input_tokens,
output_tokens = total_usage.output_tokens,
messages_count = messages.len(),
"Empty response from LLM — guard activated"
);
if any_tools_executed {
"[Task completed — the agent executed tools but did not produce a text summary.]".to_string()
} else {
"[The model returned an empty response. This usually means the model is overloaded, the context is too large, or the API key lacks credits. Try again or check /status.]".to_string()
}
}
} else {
text
Expand Down Expand Up @@ -658,6 +672,18 @@ pub async fn run_agent_loop(
consecutive_max_tokens = 0;
any_tools_executed = true;

// Capture any text content from this tool_use turn — the LLM may
// produce text alongside tool calls (e.g., a message to the user
// before calling memory_store). Without this, the text is lost if
// the next iteration returns EndTurn with empty text.
let intermediate_text = response.text();
if !intermediate_text.trim().is_empty() {
if !accumulated_text.is_empty() {
accumulated_text.push_str("\n\n");
}
accumulated_text.push_str(intermediate_text.trim());
}

// Execute tool calls
let assistant_blocks = response.content.clone();

Expand Down Expand Up @@ -1497,6 +1523,7 @@ pub async fn run_agent_loop_streaming(

let mut total_usage = TokenUsage::default();
let final_response;
let mut accumulated_text = String::new();

// Safety valve: trim excessively long message histories to prevent context overflow.
if messages.len() > MAX_HISTORY_MESSAGES {
Expand Down Expand Up @@ -1703,20 +1730,29 @@ pub async fn run_agent_loop_streaming(
}
}

// Guard against empty response — covers both iteration 0 and post-tool cycles
// Guard against empty response — use accumulated text as fallback (streaming).
let text = if text.trim().is_empty() {
warn!(
agent = %manifest.name,
iteration,
input_tokens = total_usage.input_tokens,
output_tokens = total_usage.output_tokens,
messages_count = messages.len(),
"Empty response from LLM (streaming) — guard activated"
);
if any_tools_executed {
"[Task completed — the agent executed tools but did not produce a text summary.]".to_string()
if !accumulated_text.is_empty() {
debug!(
agent = %manifest.name,
accumulated_len = accumulated_text.len(),
"Using accumulated text from intermediate tool_use iterations (streaming)"
);
accumulated_text.clone()
} else {
"[The model returned an empty response. This usually means the model is overloaded, the context is too large, or the API key lacks credits. Try again or check /status.]".to_string()
warn!(
agent = %manifest.name,
iteration,
input_tokens = total_usage.input_tokens,
output_tokens = total_usage.output_tokens,
messages_count = messages.len(),
"Empty response from LLM (streaming) — guard activated"
);
if any_tools_executed {
"[Task completed — the agent executed tools but did not produce a text summary.]".to_string()
} else {
"[The model returned an empty response. This usually means the model is overloaded, the context is too large, or the API key lacks credits. Try again or check /status.]".to_string()
}
}
} else {
text
Expand Down Expand Up @@ -1816,6 +1852,15 @@ pub async fn run_agent_loop_streaming(
consecutive_max_tokens = 0;
any_tools_executed = true;

// Capture text from intermediate tool_use turns (streaming path).
let intermediate_text = response.text();
if !intermediate_text.trim().is_empty() {
if !accumulated_text.is_empty() {
accumulated_text.push_str("\n\n");
}
accumulated_text.push_str(intermediate_text.trim());
}

let assistant_blocks = response.content.clone();

session.messages.push(Message {
Expand Down