Skip to content

fix: use instructions for system prompt and prompt for user message in subagents #6121

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
tlongwell-block merged 2 commits into from
Dec 16, 2025
Merged

fix: use instructions for system prompt and prompt for user message in subagents #6121

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
748566e
fix: use instructions for system prompt and prompt for user message i…
tlongwell-block Dec 15, 2025
f03cf38
dont double instructions if no prompt
tlongwell-block Dec 16, 2025
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
12 changes: 6 additions & 6 deletions crates/goose/src/agents/subagent_handler.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -117,11 +117,11 @@ fn get_agent_messages(
cancellation_token: Option<CancellationToken>,
) -> AgentMessagesFuture {
Box::pin(async move {
let text_instruction = recipe
.instructions
let system_instructions = recipe.instructions.clone().unwrap_or_default();
let user_task = recipe
.prompt
.clone()
.or(recipe.prompt.clone())
.ok_or_else(|| anyhow!("Recipe has no instructions or prompt"))?;
.unwrap_or_else(|| "Begin.".to_string());

let agent_manager = AgentManager::instance()
.await
Expand Down Expand Up @@ -160,7 +160,7 @@ fn get_agent_messages(
.max_turns
.expect("TaskConfig always sets max_turns"),
subagent_id: session_id.clone(),
task_instructions: text_instruction.clone(),
task_instructions: system_instructions,
tool_count: tools.len(),
available_tools: tools
.iter()
Expand All @@ -172,7 +172,7 @@ fn get_agent_messages(
.map_err(|e| anyhow!("Failed to render subagent system prompt: {}", e))?;
agent.override_system_prompt(subagent_prompt).await;

let user_message = Message::user().with_text(text_instruction);
let user_message = Message::user().with_text(user_task);
let mut conversation = Conversation::new_unvalidated(vec![user_message.clone()]);

if let Some(activities) = recipe.activities {
Expand Down
30 changes: 6 additions & 24 deletions crates/goose/src/agents/subagent_tool.rs
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -348,31 +348,13 @@ fn build_subrecipe(
)
.map_err(|e| anyhow!("Failed to build subrecipe: {}", e))?;

// Merge prompt into instructions so the subagent gets the actual task.
// The subagent handler uses `instructions` as the user message.
let mut combined = String::new();

if let Some(instructions) = &recipe.instructions {
combined.push_str(instructions);
}

if let Some(prompt) = &recipe.prompt {
if !combined.is_empty() {
combined.push_str("\n\n");
}
combined.push_str(prompt);
}

if let Some(extra_instructions) = &params.instructions {
if !combined.is_empty() {
combined.push_str("\n\n");
if let Some(extra) = &params.instructions {
let mut current = recipe.instructions.take().unwrap_or_default();
if !current.is_empty() {
current.push_str("\n\n");
}
combined.push_str("Additional context from parent agent:\n");
combined.push_str(extra_instructions);
}

if !combined.is_empty() {
recipe.instructions = Some(combined);
current.push_str(extra);
recipe.instructions = Some(current);
}

Ok(recipe)
Expand Down
2 changes: 2 additions & 0 deletions crates/goose/src/prompts/subagent_system.md
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -13,8 +13,10 @@ The maximum number of turns to respond is {{max_turns}}.
**Subagent ID**: {{subagent_id}}
{% endif %}

{% if task_instructions %}
# Task Instructions
{{task_instructions}}
{% endif %}

# Tool Usage Guidelines
**CRITICAL**: Be efficient with tool usage. Use tools only when absolutely necessary to complete your task. Here are the available tools you have access to:
Expand Down