Files
gpt-pilot/core/agents/spec_writer.py
2025-03-25 15:08:05 +01:00

210 lines
9.5 KiB
Python

from core.agents.base import BaseAgent
from core.agents.convo import AgentConvo
from core.agents.response import AgentResponse, ResponseType
from core.config import SPEC_WRITER_AGENT_NAME
from core.config.actions import SPEC_CHANGE_FEATURE_STEP_NAME, SPEC_CHANGE_STEP_NAME, SPEC_CREATE_STEP_NAME
from core.db.models import Complexity
from core.db.models.project_state import IterationStatus
from core.llm.parser import StringParser
from core.log import get_logger
from core.telemetry import telemetry
# If the project description is less than this, perform an analysis using LLM
ANALYZE_THRESHOLD = 1500
# URL to the wiki page with tips on how to write a good project description
INITIAL_PROJECT_HOWTO_URL = (
"https://github.com/Pythagora-io/gpt-pilot/wiki/How-to-write-a-good-initial-project-description"
)
log = get_logger(__name__)
class SpecWriter(BaseAgent):
agent_type = "spec-writer"
display_name = "Spec Writer"
async def run(self) -> AgentResponse:
current_iteration = self.current_state.current_iteration
if current_iteration is not None and current_iteration.get("status") == IterationStatus.NEW_FEATURE_REQUESTED:
return await self.update_spec(iteration_mode=True)
elif self.prev_response and self.prev_response.type == ResponseType.UPDATE_SPECIFICATION:
return await self.update_spec(iteration_mode=False)
else:
return await self.initialize_spec()
async def initialize_spec(self) -> AgentResponse:
self.next_state.action = SPEC_CREATE_STEP_NAME
# response = await self.ask_question(
# "Describe your app in as much detail as possible",
# allow_empty=False,
# )
# if response.cancelled:
# return AgentResponse.error(self, "No project description")
#
# user_description = response.text.strip()
user_description = self.current_state.epics[0]["description"]
complexity = await self.check_prompt_complexity(user_description)
await telemetry.trace_code_event(
"project-description",
{
"initial_prompt": user_description,
"complexity": complexity,
},
)
reviewed_spec = user_description
# if len(user_description) < ANALYZE_THRESHOLD and complexity != Complexity.SIMPLE:
# initial_spec = await self.analyze_spec(user_description)
# reviewed_spec = await self.review_spec(desc=user_description, spec=initial_spec)
self.next_state.specification = self.current_state.specification.clone()
self.next_state.specification.original_description = user_description
self.next_state.specification.description = reviewed_spec
self.next_state.specification.complexity = complexity
telemetry.set("initial_prompt", user_description)
telemetry.set("updated_prompt", reviewed_spec)
telemetry.set("is_complex_app", complexity != Complexity.SIMPLE)
return AgentResponse.done(self)
async def update_spec(self, iteration_mode) -> AgentResponse:
if iteration_mode:
self.next_state.action = SPEC_CHANGE_FEATURE_STEP_NAME
feature_description = self.current_state.current_iteration["user_feedback"]
else:
self.next_state.action = SPEC_CHANGE_STEP_NAME
feature_description = self.prev_response.data["description"]
await self.send_message(
f"Making the following changes to project specification:\n\n{feature_description}\n\nUpdated project specification:"
)
llm = self.get_llm(SPEC_WRITER_AGENT_NAME, stream_output=True)
convo = AgentConvo(self).template("add_new_feature", feature_description=feature_description)
llm_response: str = await llm(convo, temperature=0, parser=StringParser())
updated_spec = llm_response.strip()
await self.ui.generate_diff(
"project_specification", self.current_state.specification.description, updated_spec, source=self.ui_source
)
user_response = await self.ask_question(
"Do you accept these changes to the project specification?",
buttons={"yes": "Yes", "no": "No"},
default="yes",
buttons_only=True,
)
await self.ui.close_diff()
if user_response.button == "yes":
self.next_state.specification = self.current_state.specification.clone()
self.next_state.specification.description = updated_spec
telemetry.set("updated_prompt", updated_spec)
if iteration_mode:
self.next_state.current_iteration["status"] = IterationStatus.FIND_SOLUTION
self.next_state.flag_iterations_as_modified()
else:
complexity = await self.check_prompt_complexity(feature_description)
self.next_state.current_epic["complexity"] = complexity
return AgentResponse.done(self)
async def check_prompt_complexity(self, prompt: str) -> str:
is_feature = self.current_state.epics and len(self.current_state.epics) > 2
await self.send_message("Checking the complexity of the prompt ...")
llm = self.get_llm(SPEC_WRITER_AGENT_NAME)
convo = AgentConvo(self).template(
"prompt_complexity",
prompt=prompt,
is_feature=is_feature,
)
llm_response: str = await llm(convo, temperature=0, parser=StringParser())
log.info(f"Complexity check response: {llm_response}")
return llm_response.lower()
async def analyze_spec(self, spec: str) -> str:
msg = (
"Your project description seems a bit short. "
"The better you can describe the project, the better Pythagora will understand what you'd like to build.\n\n"
f"Here are some tips on how to better describe the project: {INITIAL_PROJECT_HOWTO_URL}\n\n"
"Let's start by refining your project idea:"
)
await self.send_message(msg)
llm = self.get_llm(SPEC_WRITER_AGENT_NAME, stream_output=True)
convo = AgentConvo(self).template("ask_questions").user(spec)
n_questions = 0
n_answers = 0
while True:
response: str = await llm(convo)
if len(response) > 500:
# The response is too long for it to be a question, assume it's the updated spec
confirm = await self.ask_question(
("Would you like to change or add anything? Write it out here."),
allow_empty=True,
buttons={"continue": "No thanks, the spec looks good"},
)
if confirm.cancelled or confirm.button == "continue" or confirm.text == "":
updated_spec = response.strip()
await telemetry.trace_code_event(
"spec-writer-questions",
{
"num_questions": n_questions,
"num_answers": n_answers,
"new_spec": updated_spec,
},
)
return updated_spec
convo.user(confirm.text)
else:
convo.assistant(response)
n_questions += 1
user_response = await self.ask_question(
response,
buttons={"skip": "Skip this question", "skip_all": "No more questions"},
verbose=False,
)
if user_response.cancelled or user_response.button == "skip_all":
convo.user(
"This is enough clarification, you have all the information. "
"Please output the spec now, without additional comments or questions."
)
response: str = await llm(convo)
confirm = await self.ask_question(
("Would you like to change or add anything? Write it out here."),
allow_empty=True,
buttons={"continue": "No thanks, the spec looks good"},
)
if confirm.cancelled or confirm.button == "continue" or confirm.text == "":
updated_spec = response.strip()
await telemetry.trace_code_event(
"spec-writer-questions",
{
"num_questions": n_questions,
"num_answers": n_answers,
"new_spec": updated_spec,
},
)
return updated_spec
convo.user(confirm.text)
continue
n_answers += 1
if user_response.button == "skip":
convo.user("Skip this question.")
continue
else:
convo.user(user_response.text)
async def review_spec(self, desc: str, spec: str) -> str:
convo = AgentConvo(self).template("review_spec", desc=desc, spec=spec)
llm = self.get_llm(SPEC_WRITER_AGENT_NAME)
llm_response: str = await llm(convo, temperature=0)
additional_info = llm_response.strip()
if additional_info and len(additional_info) > 6:
spec += "\n\nAdditional info/examples:\n\n" + additional_info
await self.send_message(f"\n\nAdditional info/examples:\n\n {additional_info}")
return spec