Skip to content
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

Prevent duplicate executions #1291

Merged
merged 1 commit into from
Dec 11, 2024
Merged

Prevent duplicate executions #1291

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
67 changes: 36 additions & 31 deletions agixt/Interactions.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -764,40 +764,45 @@ async def run(
# Handle commands if the prompt contains the {COMMANDS} placeholder
# We handle command injection that DOESN'T allow command execution by using {command_list} in the prompt
if "{COMMANDS}" in unformatted_prompt and "disable_commands" not in kwargs:
response = self.response
await self.execution_agent(conversation_name=conversation_name)
# While "</output>" is the ending of self.response, we need to keep executing commands or asking the agent if it should proceed
while self.response.endswith("</output>"):
response = self.response
processed_length = 0

# First handle any initial commands
if "<execute>" in self.response:
await self.execution_agent(conversation_name=conversation_name)
if self.response != response:
# There is a command output now, we need to feed it back to the LLM to continue its thought process
new_prompt = f"{formatted_prompt}\n\n{self.agent_name}: {self.response}\n\nThe assistant has executed a command and should continue its thought process, the user does not see this message. Proceed with thinking, responding, or executing more commands before the response to the user. This can be used also to evaluate output of previously executed commands and retry executing a command if the output of the command was not as expected. The assistant should never try to fill in the command output, it will be returned to the assistant after the command is executed by the system."
command_response = await self.agent.inference(
processed_length = len(self.response)

# Then enter the main processing loop
while True:
# Check if we have new commands to process
if self.response[processed_length:].strip().endswith("</output>"):
await self.execution_agent(conversation_name=conversation_name)
new_processed_length = len(self.response)

if new_processed_length > processed_length:
# Only continue if we actually got new content
new_prompt = f"{formatted_prompt}\n\n{self.agent_name}: {self.response}\n\nThe assistant has executed a command and should continue its thought process, the user does not see this message. Proceed with thinking, responding, or executing more commands before the response to the user. This can be used also to evaluate output of previously executed commands and retry executing a command if the output of the command was not as expected. The assistant should never try to fill in the command output, it will be returned to the assistant after the command is executed by the system."
command_response = await self.agent.inference(
prompt=new_prompt, tokens=tokens
)
self.response = f"{self.response}{command_response}"
processed_length = new_processed_length
else:
break # No new content, stop processing

# If no answer block yet, try to get it
elif "</answer>" not in self.response:
new_prompt = f"{formatted_prompt}\n\n{self.agent_name}: {self.response}\n\nWas the assistant {self.agent_name} done typing? If not, continue from where you left off without acknowledging this message or repeating anything that was already typed and the response will be appended. If the assistant needs to rewrite the response, start a new <answer> tag with the new response and close it with </answer> when complete. If the assistant was done, simply respond with '</answer>.' to send the message to the user."
response = await self.agent.inference(
prompt=new_prompt, tokens=tokens
)
self.response = f"{self.response}{command_response}"
if "<answer>" in formatted_prompt and "</answer>" not in self.response:
while "</answer>" not in self.response:
new_prompt = f"{formatted_prompt}\n\n{self.agent_name}: {self.response}\n\nWas the assistant {self.agent_name} done typing? If not, continue from where you left off without acknowledging this message or repeating anything that was already typed and the response will be appended. If the assistant needs to rewrite the response, start a new <answer> tag with the new response and close it with </answer> when complete. If the assistant was done, simply respond with '</answer>.' to send the message to the user."
response = await self.agent.inference(prompt=new_prompt, tokens=tokens)
self.response = f"{self.response}{response}"
if (
"</answer>" not in self.response
and "{COMMANDS}" in unformatted_prompt
and "disable_commands" not in kwargs
):
await self.execution_agent(conversation_name=conversation_name)
while self.response.endswith("</output>"):
response = self.response
await self.execution_agent(conversation_name=conversation_name)
if self.response != response:
# There is a command output now, we need to feed it back to the LLM to continue its thought process
new_prompt = f"{formatted_prompt}\n\n{self.agent_name}: {self.response}\n\nThe assistant has executed a command and should continue its thought process, the user does not see this message. Proceed with thinking, responding, or executing more commands before the response to the user. This can be used also to evaluate output of previously executed commands and retry executing a command if the output of the command was not as expected. The assistant should never try to fill in the command output, it will be returned to the assistant after the command is executed by the system."
command_response = await self.agent.inference(
prompt=new_prompt, tokens=tokens
)
self.response = f"{self.response}{command_response}"
self.response = f"{self.response}{response}"

# After getting more response, let the loop continue to check for any new commands
continue

else:
# We have an answer block and no new commands to process
break
if self.response != "" and self.response != None:
agent_settings = self.agent.AGENT_CONFIG["settings"]
if "<audio controls>" in self.response:
Expand Down
Loading