Ling0925
/
suna
mirror of https://github.com/kortix-ai/suna.git
1
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

fix error where it wont add messages to thread if max_xml_tool_calls x > 0

This commit is contained in:
marko-kraemer 2025-04-10 18:13:12 +01:00
parent 329d635dfa
commit 700ccddbad
6 changed files with 58 additions and 51 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

25
backend/agent/prompt.py
View File

@ -53,8 +53,6 @@ TODO.MD is your central planning tool and source of truth for all tasks. It driv
- Begin with 5-10 major sections covering the entire task lifecycle
- Include thorough preparation and research sections before implementation
- Format as markdown checklist with clear, actionable items: `- [ ] Task description`
- Include current timestamp and task ID for tracking
- Add estimated completion time for each section
- Build a complete roadmap before starting execution
2. SECTION-BASED PROGRESSION: Work on one complete section at a time:
@ -73,8 +71,6 @@ TODO.MD is your central planning tool and source of truth for all tasks. It driv
- Carefully evaluate the results before proceeding
- Mark completed items with `- [x]` using text replacement
- Add new discovered subtasks as needed
- Update task progress estimates
- Add timestamps to completed items
- Document observations and learnings
5. PROGRESSION GATES: Never advance to a new section until:
@ -88,7 +84,6 @@ TODO.MD is your central planning tool and source of truth for all tasks. It driv
- Preserve completed tasks with their status
- Add, modify or remove pending tasks
- Document reason for changes in todo.md
- Re-estimate completion times
- Ensure the modified plan maintains logical progression
Always reference todo.md by line number when making decisions or reporting progress.
@ -115,10 +110,9 @@ You operate in a methodical, single-step agent loop guided by todo.md:
- No further actions until execution completes
4. PROGRESS TRACKING: Update todo.md with detailed progress:
- Mark completed items with timestamps
- Mark completed items
- Add new discovered tasks as needed
- Document lessons learned and observations
- Update estimates for remaining work
5. METHODICAL ITERATION: Repeat steps 1-4 until section completion:
- Choose only one tool call per iteration
@ -161,7 +155,7 @@ The planner module is responsible for initializing and organizing your todo.md w
<todo_format>
Todo.md must follow this comprehensive structured format with many sections:
```
# Task: [Task Name] - Created [Timestamp]
# Task: [Task Name]
## 1. Task Analysis and Planning
- [ ] 1.1 Understand user requirements completely
@ -169,7 +163,6 @@ Todo.md must follow this comprehensive structured format with many sections:
- [ ] 1.3 Research similar existing solutions
- [ ] 1.4 Define success criteria and deliverables
- [ ] 1.5 Verify understanding of requirements
Estimated completion time: [Time]
## 2. Environment Setup and Preparation
- [ ] 2.1 Check current environment state
@ -177,7 +170,6 @@ Estimated completion time: [Time]
- [ ] 2.3 Set up project structure
- [ ] 2.4 Configure development tools
- [ ] 2.5 Verify environment readiness
Estimated completion time: [Time]
## 3. Research and Information Gathering
- [ ] 3.1 Search for relevant documentation
@ -185,7 +177,6 @@ Estimated completion time: [Time]
- [ ] 3.3 Collect reference materials
- [ ] 3.4 Organize findings
- [ ] 3.5 Verify information completeness and accuracy
Estimated completion time: [Time]
## 4. Design and Architecture
- [ ] 4.1 Create system architecture diagram
@ -193,7 +184,6 @@ Estimated completion time: [Time]
- [ ] 4.3 Design data structures
- [ ] 4.4 Plan implementation approach
- [ ] 4.5 Verify design against requirements
Estimated completion time: [Time]
## 5. Implementation - Component A
- [ ] 5.1 Implement core functionality
@ -201,7 +191,6 @@ Estimated completion time: [Time]
- [ ] 5.3 Optimize performance
- [ ] 5.4 Document code
- [ ] 5.5 Verify component functionality
Estimated completion time: [Time]
## 6. Implementation - Component B
- [ ] 6.1 Implement core functionality
@ -209,7 +198,6 @@ Estimated completion time: [Time]
- [ ] 6.3 Optimize performance
- [ ] 6.4 Document code
- [ ] 6.5 Verify component functionality
Estimated completion time: [Time]
## 7. Integration and Testing
- [ ] 7.1 Integrate all components
@ -217,7 +205,6 @@ Estimated completion time: [Time]
- [ ] 7.3 Fix identified issues
- [ ] 7.4 Verify system behavior
- [ ] 7.5 Document test results
Estimated completion time: [Time]
## 8. Deployment and Delivery
- [ ] 8.1 Prepare deployment package
@ -225,7 +212,6 @@ Estimated completion time: [Time]
- [ ] 8.3 Verify deployment success
- [ ] 8.4 Document deployment process
- [ ] 8.5 Prepare user documentation
Estimated completion time: [Time]
## 9. Final Verification
- [ ] 9.1 Validate all deliverables against requirements
@ -233,14 +219,13 @@ Estimated completion time: [Time]
- [ ] 9.3 Prepare comprehensive summary
- [ ] 9.4 Compile all documentation
- [ ] 9.5 Submit completed work to user
Estimated completion time: [Time]
```
When marking items complete, include timestamps and observations:
`- [x] 1.1 Understand user requirements completely - Completed [Timestamp] - [Brief observation]`
When marking items complete, include observations:
`- [x] 1.1 Understand user requirements completely - [Brief observation]`
SECTION TRANSITIONS must be documented:
`## Completed Section: [Section Name] - [Timestamp]
`## Completed Section: [Section Name]
Summary: [Comprehensive summary of section achievements and insights]`
</todo_format>

26
backend/agent/run.py
View File

@ -9,7 +9,7 @@ from agentpress.thread_manager import ThreadManager
from agentpress.response_processor import ProcessorConfig
from agent.tools.sb_browse_tool import SandboxBrowseTool
from agent.tools.sb_shell_tool import SandboxShellTool
from agent.tools.sb_website_tool import SandboxWebsiteTool
# from agent.tools.sb_website_tool import SandboxWebsiteTool
from agent.tools.sb_files_tool import SandboxFilesTool
from agent.prompt import get_system_prompt
from agent.tools.utils.daytona_sandbox import daytona, create_sandbox, get_or_start_sandbox
@ -28,7 +28,7 @@ async def run_agent(thread_id: str, project_id: str, stream: bool = True, thread
if project.data[0]['sandbox_id']:
sandbox_id = project.data[0]['sandbox_id']
sandbox_pass = project.data[0]['sandbox_pass']
sandbox = await get_or_start_sandbox(sandbox_id, sandbox_pass)
sandbox = await get_or_start_sandbox(sandbox_id)
else:
sandbox_pass = str(uuid4())
sandbox = create_sandbox(sandbox_pass)
@ -37,13 +37,13 @@ async def run_agent(thread_id: str, project_id: str, stream: bool = True, thread
'sandbox_id': sandbox_id,
'sandbox_pass': sandbox_pass
}).eq('project_id', project_id).execute()
### ---
thread_manager.add_tool(SandboxBrowseTool, sandbox_id=sandbox_id, password=sandbox_pass)
thread_manager.add_tool(SandboxWebsiteTool, sandbox_id=sandbox_id, password=sandbox_pass)
# thread_manager.add_tool(SandboxWebsiteTool, sandbox_id=sandbox_id, password=sandbox_pass)
thread_manager.add_tool(SandboxShellTool, sandbox_id=sandbox_id, password=sandbox_pass)
thread_manager.add_tool(SandboxFilesTool, sandbox_id=sandbox_id, password=sandbox_pass)
files_tool = SandboxFilesTool(sandbox_id=sandbox_id, password=sandbox_pass)
system_message = { "role": "system", "content": get_system_prompt() }
@ -56,8 +56,6 @@ async def run_agent(thread_id: str, project_id: str, stream: bool = True, thread
# model_name = "bedrock/anthropic.claude-3-7-sonnet-20250219-v1:0"
# model_name = "bedrock/anthropic.claude-3-5-sonnet-20241022-v2:0"
files_tool = SandboxFilesTool(sandbox_id=sandbox_id, password=sandbox_pass)
iteration_count = 0
continue_execution = True
@ -73,13 +71,15 @@ async def run_agent(thread_id: str, project_id: str, stream: bool = True, thread
state_message = {
"role": "user",
"content": f"""
Current development environment workspace state:
Current workspace state:
<current_workspace_state>
{state_str}
</current_workspace_state>
"""
}
# print(f"State message: {state_message}")
response = await thread_manager.run_thread(
thread_id=thread_id,
system_prompt=system_message,
@ -91,8 +91,8 @@ Current development environment workspace state:
tool_choice="auto",
max_xml_tool_calls=1,
processor_config=ProcessorConfig(
xml_tool_calling=False,
native_tool_calling=True,
xml_tool_calling=True,
native_tool_calling=False,
execute_tools=True,
execute_on_stream=True,
tool_execution_strategy="parallel",
@ -116,6 +116,13 @@ Current development environment workspace state:
function_name = tool_call.get('function', {}).get('name', '')
if function_name in ['message_ask_user', 'idle']:
last_tool_call = function_name
# Check for XML versions like <message_ask_user> or <Idle> in content chunks
elif chunk.get('type') == 'content' and 'content' in chunk:
content = chunk.get('content', '')
if '<message_ask_user>' in content or '<Idle>' in content:
xml_tool = 'message_ask_user' if '<message_ask_user>' in content else 'idle'
last_tool_call = xml_tool
print(f"Agent used XML tool: {xml_tool}")
yield chunk
@ -126,7 +133,6 @@ Current development environment workspace state:
# TESTING
async def test_agent():

22
backend/agent/tools/sb_files_tool.py
View File

@ -94,11 +94,10 @@ class SandboxFilesTool(SandboxToolsBase):
tag_name="create-file",
mappings=[
{"param_name": "file_path", "node_type": "attribute", "path": "."},
{"param_name": "file_contents", "node_type": "content", "path": "."},
{"param_name": "permissions", "node_type": "attribute", "path": "@permissions"}
{"param_name": "file_contents", "node_type": "content", "path": "."}
],
example='''
<create-file file_path="path/to/file" permissions="644">
<create-file file_path="path/to/file">
File contents go here
</create-file>
'''
@ -225,11 +224,10 @@ class SandboxFilesTool(SandboxToolsBase):
tag_name="full-file-rewrite",
mappings=[
{"param_name": "file_path", "node_type": "attribute", "path": "."},
{"param_name": "file_contents", "node_type": "content", "path": "."},
{"param_name": "permissions", "node_type": "attribute", "path": "@permissions"}
{"param_name": "file_contents", "node_type": "content", "path": "."}
],
example='''
<full-file-rewrite file_path="path/to/file" permissions="644">
<full-file-rewrite file_path="path/to/file">
New file contents go here, replacing all existing content
</full-file-rewrite>
'''
@ -316,9 +314,9 @@ class SandboxFilesTool(SandboxToolsBase):
@xml_schema(
tag_name="search-files",
mappings=[
{"param_name": "path", "node_type": "attribute", "path": "@path"},
{"param_name": "pattern", "node_type": "attribute", "path": "@pattern"},
{"param_name": "recursive", "node_type": "attribute", "path": "@recursive"}
{"param_name": "path", "node_type": "attribute", "path": "."},
{"param_name": "pattern", "node_type": "attribute", "path": "."},
{"param_name": "recursive", "node_type": "attribute", "path": "."}
],
example='''
<search-files path="path/to/search" pattern="text-of-interest" recursive="true">
@ -378,9 +376,9 @@ class SandboxFilesTool(SandboxToolsBase):
@xml_schema(
tag_name="replace-in-file",
mappings=[
{"param_name": "file", "node_type": "attribute", "path": "@file"},
{"param_name": "pattern", "node_type": "element", "path": "pattern"},
{"param_name": "new_value", "node_type": "element", "path": "new_value"}
{"param_name": "file", "node_type": "attribute", "path": "."},
{"param_name": "pattern", "node_type": "element", "path": "."},
{"param_name": "new_value", "node_type": "element", "path": "."}
],
example='''
<replace-in-file file="path/to/file.txt">

2
backend/agent/tools/utils/daytona_sandbox.py
View File

@ -270,7 +270,7 @@ def wait_for_api_ready(sandbox):
return api_url
async def get_or_start_sandbox(sandbox_id: str, sandbox_pass: str):
async def get_or_start_sandbox(sandbox_id: str):
"""Retrieve a sandbox by ID, check its state, and start it if needed.
Also ensure the sandbox_browser_api and HTTP server services are running."""

32
backend/agentpress/response_processor.py
View File

@ -81,7 +81,10 @@ class ResponseProcessor:
Args:
tool_registry: Registry of available tools
add_message_callback: Callback function to add messages to the thread
add_message_callback: Callback function to add messages to the thread.
This function is used to record assistant messages, tool calls,
and tool results in the conversation history, making them
available for the LLM in subsequent interactions.
"""
self.tool_registry = tool_registry
self.add_message = add_message_callback
@ -426,8 +429,8 @@ class ResponseProcessor:
logger.info(f"Stream finished with reason: xml_tool_limit_reached after {xml_tool_call_count} XML tool calls")
# After streaming completes, process any remaining content and tool calls
# Only do this if we didn't stop due to XML tool limit (already handled pending executions above)
if accumulated_content and finish_reason != "xml_tool_limit_reached":
# IMPORTANT: Always process accumulated content even when XML tool limit is reached
if accumulated_content:
# Extract final complete tool calls for native format
complete_native_tool_calls = []
if config.native_tool_calling:
@ -621,7 +624,7 @@ class ResponseProcessor:
}
})
# Add assistant message FIRST
# Add assistant message FIRST - always do this regardless of finish_reason
message_data = {
"role": "assistant",
"content": content,
@ -1088,7 +1091,20 @@ class ResponseProcessor:
result: ToolResult,
strategy: Union[XmlAddingStrategy, str] = "assistant_message"
):
"""Add a tool result to the thread based on the specified format."""
"""Add a tool result to the conversation thread based on the specified format.
This method formats tool results and adds them to the conversation history,
making them visible to the LLM in subsequent interactions. Results can be
added either as native tool messages (OpenAI format) or as XML-wrapped content
with a specified role (user or assistant).
Args:
thread_id: ID of the conversation thread
tool_call: The original tool call that produced this result
result: The result from the tool execution
strategy: How to add XML tool results to the conversation
("user_message", "assistant_message", or "inline_edit")
"""
try:
# Check if this is a native function call (has id field)
if "id" in tool_call:
@ -1122,7 +1138,8 @@ class ResponseProcessor:
logger.info(f"Adding native tool result for tool_call_id={tool_call['id']} with role=tool")
# Add as a tool message
# Add as a tool message to the conversation history
# This makes the result visible to the LLM in the next turn
await self.add_message(
thread_id=thread_id,
type="tool", # Special type for tool responses
@ -1142,7 +1159,8 @@ class ResponseProcessor:
# Format the content using the formatting helper
content = self._format_xml_tool_result(tool_call, result)
# Add the message with the appropriate role
# Add the message with the appropriate role to the conversation history
# This allows the LLM to see the tool result in subsequent interactions
result_message = {
"role": result_role,
"content": content

2
backend/utils/logger.py
View File

@ -83,7 +83,7 @@ def setup_logger(name: str = 'agentpress') -> logging.Logger:
# Console handler
console_handler = logging.StreamHandler(sys.stdout)
console_handler.setLevel(logging.INFO)
console_handler.setLevel(logging.DEBUG)
# Create formatters
file_formatter = logging.Formatter(