merge main

core/agents/developer.py

@@ -8,6 +8,7 @@ from core.agents.base import BaseAgent
 from core.agents.convo import AgentConvo
 from core.agents.mixins import RelevantFilesMixin
 from core.agents.response import AgentResponse, ResponseType
+from core.config import TASK_BREAKDOWN_AGENT_NAME
 from core.db.models.project_state import IterationStatus, TaskStatus
 from core.db.models.specification import Complexity
 from core.llm.parser import JSONParser
@@ -216,7 +217,7 @@ class Developer(RelevantFilesMixin, BaseAgent):
 
         current_task_index = self.current_state.tasks.index(current_task)
 
-        llm = self.get_llm()
+        llm = self.get_llm(TASK_BREAKDOWN_AGENT_NAME)
         convo = AgentConvo(self).template(
             "breakdown",
             task=current_task,
@@ -234,6 +235,7 @@ class Developer(RelevantFilesMixin, BaseAgent):
         }
         self.next_state.flag_tasks_as_modified()
 
+        llm = self.get_llm()
         await self.send_message("Breaking down the task into steps ...")
         convo.assistant(response).template("parse_task").require_schema(TaskSteps)
         response: TaskSteps = await llm(convo, parser=JSONParser(TaskSteps), temperature=0)

core/agents/error_handler.py

@@ -111,7 +111,7 @@ class ErrorHandler(BaseAgent):
                 "description": llm_response,
                 "alternative_solutions": [],
                 "attempts": 1,
-                "status": IterationStatus.HUNTING_FOR_BUG,
+                "status": IterationStatus.IMPLEMENT_SOLUTION,
                 "bug_hunting_cycles": [],
             }
         ]

core/agents/orchestrator.py

@@ -182,6 +182,8 @@ class Orchestrator(BaseAgent):
                 return Importer(self.state_manager, self.ui, prev_response=prev_response)
             if prev_response.type == ResponseType.EXTERNAL_DOCS_REQUIRED:
                 return ExternalDocumentation(self.state_manager, self.ui, prev_response=prev_response)
+            if prev_response.type == ResponseType.UPDATE_SPECIFICATION:
+                return SpecWriter(self.state_manager, self.ui, prev_response=prev_response)
 
         if not state.specification.description:
             if state.files:
@@ -252,6 +254,9 @@ class Orchestrator(BaseAgent):
             elif current_iteration_status == IterationStatus.PROBLEM_SOLVER:
                 # Call Problem Solver if the user said "I'm stuck in a loop"
                 return ProblemSolver(self.state_manager, self.ui)
+            elif current_iteration_status == IterationStatus.NEW_FEATURE_REQUESTED:
+                # Call Spec Writer to add the "change" requested by the user to project specification
+                return SpecWriter(self.state_manager, self.ui)
 
         # We have just finished the task, call Troubleshooter to ask the user to review
         return Troubleshooter(self.state_manager, self.ui)

core/agents/response.py

@@ -45,6 +45,9 @@ class ResponseType(str, Enum):
     EXTERNAL_DOCS_REQUIRED = "external-docs-required"
     """We need to fetch external docs for a task."""
 
+    UPDATE_SPECIFICATION = "update-specification"
+    """We need to update the project specification."""
+
 
 class AgentResponse:
     type: ResponseType = ResponseType.DONE
@@ -144,3 +147,13 @@ class AgentResponse:
     @staticmethod
     def external_docs_required(agent: "BaseAgent") -> "AgentResponse":
         return AgentResponse(type=ResponseType.EXTERNAL_DOCS_REQUIRED, agent=agent)
+
+    @staticmethod
+    def update_specification(agent: "BaseAgent", description: str) -> "AgentResponse":
+        return AgentResponse(
+            type=ResponseType.UPDATE_SPECIFICATION,
+            agent=agent,
+            data={
+                "description": description,
+            },
+        )

core/agents/spec_writer.py

@@ -1,7 +1,8 @@
 from core.agents.base import BaseAgent
 from core.agents.convo import AgentConvo
-from core.agents.response import AgentResponse
+from core.agents.response import AgentResponse, ResponseType
 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
@@ -26,6 +27,15 @@ class SpecWriter(BaseAgent):
     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:
         response = await self.ask_question(
             "Describe your app in as much detail as possible",
             allow_empty=False,
@@ -67,6 +77,7 @@ class SpecWriter(BaseAgent):
             spec = await self.review_spec(spec)
 
         self.next_state.specification = self.current_state.specification.clone()
+        self.next_state.specification.original_description = spec
         self.next_state.specification.description = spec
         self.next_state.specification.complexity = complexity
         telemetry.set("initial_prompt", spec)
@@ -75,6 +86,42 @@ class SpecWriter(BaseAgent):
         self.next_state.action = SPEC_STEP_NAME
         return AgentResponse.done(self)
 
+    async def update_spec(self, iteration_mode) -> AgentResponse:
+        if iteration_mode:
+            feature_description = self.current_state.current_iteration["user_feedback"]
+        else:
+            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()
+        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(self.current_state.specification.description, updated_spec)
+        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(user_response.text)
+            self.next_state.current_epic["complexity"] = complexity
+
+        return AgentResponse.done(self)
+
     async def check_prompt_complexity(self, prompt: str) -> str:
         await self.send_message("Checking the complexity of the prompt ...")
         llm = self.get_llm()
@@ -160,6 +207,6 @@ class SpecWriter(BaseAgent):
         llm = self.get_llm()
         llm_response: str = await llm(convo, temperature=0)
         additional_info = llm_response.strip()
-        if additional_info:
+        if additional_info and len(additional_info) > 6:
             spec += "\nAdditional info/examples:\n" + additional_info
         return spec

core/agents/task_reviewer.py

@@ -47,7 +47,7 @@ class TaskReviewer(BaseAgent):
         )
         llm_response: str = await llm(convo, temperature=0.7)
 
-        if "done" in llm_response.strip().lower()[-7:]:
+        if "done" in llm_response.strip().lower()[-20:]:
             return AgentResponse.done(self)
         else:
             return AgentResponse.task_review_feedback(self, llm_response)

core/agents/tech_lead.py

@@ -5,7 +5,6 @@ from pydantic import BaseModel, Field
 from core.agents.base import BaseAgent
 from core.agents.convo import AgentConvo
 from core.agents.response import AgentResponse
-from core.db.models import Complexity
 from core.db.models.project_state import TaskStatus
 from core.llm.parser import JSONParser
 from core.log import get_logger
@@ -107,7 +106,8 @@ class TechLead(BaseAgent):
 
         if summaries:
             spec = self.current_state.specification.clone()
-            spec.description += "\n\n" + "\n\n".join(summaries)
+            spec.template_summary = "\n\n".join(summaries)
+
             self.next_state.specification = spec
 
     async def ask_for_new_feature(self) -> AgentResponse:
@@ -135,12 +135,12 @@ class TechLead(BaseAgent):
                 "description": response.text,
                 "summary": None,
                 "completed": False,
-                "complexity": Complexity.HARD,
+                "complexity": None,  # Determined and defined in SpecWriter
             }
         ]
         # Orchestrator will rerun us to break down the new feature epic
         self.next_state.action = f"Start of feature #{len(self.current_state.epics)}"
-        return AgentResponse.done(self)
+        return AgentResponse.update_specification(self, response.text)
 
     async def plan_epic(self, epic) -> AgentResponse:
         log.debug(f"Planning tasks for the epic: {epic['name']}")

core/agents/troubleshooter.py

@@ -87,7 +87,7 @@ class Troubleshooter(IterationPromptMixin, RelevantFilesMixin, BaseAgent):
             return await self.complete_task()
 
         user_feedback = bug_report or change_description
-        user_feedback_qa = await self.generate_bug_report(run_command, user_instructions, user_feedback)
+        user_feedback_qa = None  # await self.generate_bug_report(run_command, user_instructions, user_feedback)
 
         if is_loop:
             if last_iteration["alternative_solutions"]:
@@ -103,14 +103,14 @@ class Troubleshooter(IterationPromptMixin, RelevantFilesMixin, BaseAgent):
             # should be - elif change_description is not None: - but to prevent bugs with the extension
             # this might be caused if we show the input field instead of buttons
             await self.get_relevant_files(user_feedback)
-            iteration_status = IterationStatus.FIND_SOLUTION
+            iteration_status = IterationStatus.NEW_FEATURE_REQUESTED
 
         self.next_state.iterations = self.current_state.iterations + [
             {
                 "id": uuid4().hex,
                 "user_feedback": user_feedback,
                 "user_feedback_qa": user_feedback_qa,
-                "description": change_description,
+                "description": None,
                 "alternative_solutions": [],
                 # FIXME - this is incorrect if this is a new problem; otherwise we could
                 # just count the iterations
@@ -224,7 +224,7 @@ class Troubleshooter(IterationPromptMixin, RelevantFilesMixin, BaseAgent):
         is_loop = False
         should_iterate = True
 
-        test_message = "Can you check if the app works please?"
+        test_message = "Please check if the app is working"
         if user_instructions:
             hint = " Here is a description of what should be working:\n\n" + user_instructions
 
@@ -260,13 +260,11 @@ class Troubleshooter(IterationPromptMixin, RelevantFilesMixin, BaseAgent):
             is_loop = True
 
         elif user_response.button == "change":
-            user_description = await self.ask_question(
-                "Please describe the change you want to make (one at the time please)"
-            )
+            user_description = await self.ask_question("Please describe the change you want to make (one at a time)")
             change_description = user_description.text
 
         elif user_response.button == "bug":
-            user_description = await self.ask_question("Please describe the issue you found (one at the time please)")
+            user_description = await self.ask_question("Please describe the issue you found (one at a time)")
             bug_report = user_description.text
 
         return should_iterate, is_loop, bug_report, change_description

core/config/__init__.py

@@ -37,6 +37,7 @@ DEFAULT_AGENT_NAME = "default"
 CODE_MONKEY_AGENT_NAME = "CodeMonkey"
 DESCRIBE_FILES_AGENT_NAME = "CodeMonkey.describe_files"
 CHECK_LOGS_AGENT_NAME = "BugHunter.check_logs"
+TASK_BREAKDOWN_AGENT_NAME = "Developer.breakdown_current_task"
 
 # Endpoint for the external documentation
 EXTERNAL_DOCUMENTATION_API = "http://docs-pythagora-io-439719575.us-east-1.elb.amazonaws.com"
@@ -317,7 +318,8 @@ class Config(_StrictModel):
             DEFAULT_AGENT_NAME: AgentLLMConfig(),
             CODE_MONKEY_AGENT_NAME: AgentLLMConfig(model="gpt-4-0125-preview", temperature=0.0),
             DESCRIBE_FILES_AGENT_NAME: AgentLLMConfig(model="gpt-3.5-turbo", temperature=0.0),
-            CHECK_LOGS_AGENT_NAME: AgentLLMConfig(model="claude-3-5-sonnet-20240620", temperature=0.0),
+            CHECK_LOGS_AGENT_NAME: AgentLLMConfig(model="claude-3-5-sonnet-20240620", temperature=0.5),
+            TASK_BREAKDOWN_AGENT_NAME: AgentLLMConfig(model="claude-3-5-sonnet-20240620", temperature=0.5),
         }
     )
     prompt: PromptConfig = PromptConfig()

core/db/migrations/versions/c8905d4ce784_add_original_description_and_template_.py

@@ -0,0 +1,36 @@
+"""Add original description and template summary fields to specifications
+
+Revision ID: c8905d4ce784
+Revises: 08d71952ec2f
+Create Date: 2024-07-25 19:24:23.808237
+
+"""
+
+from typing import Sequence, Union
+
+import sqlalchemy as sa
+from alembic import op
+
+# revision identifiers, used by Alembic.
+revision: str = "c8905d4ce784"
+down_revision: Union[str, None] = "08d71952ec2f"
+branch_labels: Union[str, Sequence[str], None] = None
+depends_on: Union[str, Sequence[str], None] = None
+
+
+def upgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    with op.batch_alter_table("specifications", schema=None) as batch_op:
+        batch_op.add_column(sa.Column("original_description", sa.String(), nullable=True))
+        batch_op.add_column(sa.Column("template_summary", sa.String(), nullable=True))
+
+    # ### end Alembic commands ###
+
+
+def downgrade() -> None:
+    # ### commands auto generated by Alembic - please adjust! ###
+    with op.batch_alter_table("specifications", schema=None) as batch_op:
+        batch_op.drop_column("template_summary")
+        batch_op.drop_column("original_description")
+
+    # ### end Alembic commands ###

core/db/models/project_state.py

@@ -41,6 +41,7 @@ class IterationStatus:
     IMPLEMENT_SOLUTION = "implement_solution"
     FIND_SOLUTION = "find_solution"
     PROBLEM_SOLVER = "problem_solver"
+    NEW_FEATURE_REQUESTED = "new_feature_requested"
     DONE = "done"
 
 

core/db/models/specification.py

@@ -26,7 +26,9 @@ class Specification(Base):
     id: Mapped[int] = mapped_column(primary_key=True, autoincrement=True)
 
     # Attributes
+    original_description: Mapped[Optional[str]] = mapped_column()
     description: Mapped[str] = mapped_column(default="")
+    template_summary: Mapped[Optional[str]] = mapped_column()
     architecture: Mapped[str] = mapped_column(default="")
     system_dependencies: Mapped[list[dict]] = mapped_column(default=list)
     package_dependencies: Mapped[list[dict]] = mapped_column(default=list)
@@ -43,7 +45,9 @@ class Specification(Base):
         Clone the specification.
         """
         clone = Specification(
+            original_description=self.original_description,
             description=self.description,
+            template_summary=self.template_summary,
             architecture=self.architecture,
             system_dependencies=self.system_dependencies,
             package_dependencies=self.package_dependencies,

core/llm/anthropic_client.py

@@ -15,6 +15,7 @@ log = get_logger(__name__)
 
 # Maximum number of tokens supported by Anthropic Claude 3
 MAX_TOKENS = 4096
+MAX_TOKENS_SONNET = 8192
 
 
 class AnthropicClient(BaseLLMClient):
@@ -72,6 +73,11 @@ class AnthropicClient(BaseLLMClient):
             "messages": messages,
             "temperature": self.config.temperature if temperature is None else temperature,
         }
+
+        if "sonnet" in self.config.model:
+            completion_kwargs["extra_headers"] = {"anthropic-beta": "max-tokens-3-5-sonnet-2024-07-15"}
+            completion_kwargs["max_tokens"] = MAX_TOKENS_SONNET
+
         if json_mode:
             completion_kwargs["response_format"] = {"type": "json_object"}
 

core/prompts/developer/breakdown.prompt

@@ -4,21 +4,6 @@ You are working on an app called "{{ state.branch.project.name }}" and you need
 {% include "partials/features_list.prompt" %}
 {% include "partials/files_list.prompt" %}
 
-We've broken the development of this {% if state.epics|length > 1 %}feature{% else %}app{% endif %} down to these tasks:
-```
-{% for task in state.tasks %}
-{{ loop.index }}. {{ task.description }}{% if task.get("status") == "done" %} (completed){% endif %}
-{% endfor %}
-```
-
-You are currently working on task #{{ current_task_index + 1 }} with the following description:
-```
-{{ task.description }}
-```
-{% if current_task_index != 0 %}All previous tasks are finished and you don't have to work on them.{% endif %}
-
-Now, tell me all the code that needs to be written to implement ONLY this task and have it fully working and all commands that need to be run to implement this task.
-
 {% include "partials/doc_snippets.prompt" %}
 
 {%- if state.epics|length == 1 %}
@@ -35,3 +20,19 @@ DO NOT specify commands to create any folders or files, they will be created aut
 {% include "partials/file_size_limit.prompt" %}
 
 Never use the port 5000 to run the app, it's reserved.
+
+--IMPLEMENTATION INSTRUCTIONS--
+We've broken the development of this {% if state.epics|length > 1 %}feature{% else %}app{% endif %} down to these tasks:
+```
+{% for task in state.tasks %}
+{{ loop.index }}. {{ task.description }}{% if task.get("status") == "done" %} (completed){% endif %}
+{% endfor %}
+```
+
+You are currently working on task #{{ current_task_index + 1 }} with the following description:
+```
+{{ task.description }}
+```
+{% if current_task_index != 0 %}All previous tasks are finished and you don't have to work on them.{% endif %}
+
+Now, tell me all the code that needs to be written to implement ONLY this task and have it fully working and all commands that need to be run to implement this task.

core/prompts/spec-writer/add_new_feature.prompt

@@ -0,0 +1,19 @@
+Your team has taken the client brief and turned it into a project specification.
+Afterwards the client added a description for a new feature to be added to the project specification.
+Your job is to update the project specification so that it contains the new feature information but does not lack any of the information from the original project specification.
+
+This might include:
+* details on how the app should work
+* information which 3rd party packages or APIs to use or avoid
+* concrete examples of API requests/responses, library usage, or other external documentation
+
+Here is the original project specification:
+{{ state.specification.description }}
+
+Here is the new feature description:
+---FEATURE-DESCRIPTION-START---
+{{ feature_description }}
+---FEATURE-DESCRIPTION-END---
+
+In your response, output only the new updated project specification, without any additional messages to the user. 
+If there is no feature description just output the original project specification.

core/telemetry/__init__.py

@@ -82,6 +82,8 @@ class Telemetry:
             "model": config.agent["default"].model,
             # Initial prompt
             "initial_prompt": None,
+            # Updated prompt
+            "updated_prompt": None,
             # App complexity
             "is_complex_app": None,
             # Optional template used for the project

core/templates/registry.py

@@ -2,8 +2,10 @@ from enum import Enum
 
 from core.log import get_logger
 
+from .javascript_react import JavascriptReactProjectTemplate
 from .node_express_mongoose import NodeExpressMongooseProjectTemplate
-from .react_express import ReactExpressProjectTemplate
+
+# from .react_express import ReactExpressProjectTemplate
 
 log = get_logger(__name__)
 
@@ -11,13 +13,13 @@ log = get_logger(__name__)
 class ProjectTemplateEnum(str, Enum):
     """Choices of available project templates."""
 
-    # JAVASCRIPT_REACT = JavascriptReactProjectTemplate.name
+    JAVASCRIPT_REACT = JavascriptReactProjectTemplate.name
     NODE_EXPRESS_MONGOOSE = NodeExpressMongooseProjectTemplate.name
-    REACT_EXPRESS = ReactExpressProjectTemplate.name
+    # REACT_EXPRESS = ReactExpressProjectTemplate.name
 
 
 PROJECT_TEMPLATES = {
-    # JavascriptReactProjectTemplate.name: JavascriptReactProjectTemplate,
+    JavascriptReactProjectTemplate.name: JavascriptReactProjectTemplate,
     NodeExpressMongooseProjectTemplate.name: NodeExpressMongooseProjectTemplate,
-    ReactExpressProjectTemplate.name: ReactExpressProjectTemplate,
+    # ReactExpressProjectTemplate.name: ReactExpressProjectTemplate,
 }

core/ui/ipc_client.py

@@ -42,6 +42,8 @@ class MessageType(str, Enum):
     IMPORT_PROJECT = "importProject"
     APP_FINISHED = "appFinished"
     FEATURE_FINISHED = "featureFinished"
+    GENERATE_DIFF = "generateDiff"
+    CLOSE_DIFF = "closeDiff"
 
 
 class Message(BaseModel):
@@ -356,6 +358,19 @@ class IPCClientUI(UIBase):
             content=stats,
         )
 
+    async def generate_diff(self, file_old: str, file_new: str):
+        await self._send(
+            MessageType.GENERATE_DIFF,
+            content={
+                "file_old": file_old,
+                "file_new": file_new,
+            },
+        )
+
+    async def close_diff(self):
+        log.debug("Sending signal to close the generated diff file")
+        await self._send(MessageType.CLOSE_DIFF)
+
     async def loading_finished(self):
         log.debug("Sending project loading finished signal to the extension")
         await self._send(MessageType.LOADING_FINISHED)

tests/agents/test_tech_lead.py

@@ -56,7 +56,7 @@ async def test_ask_for_feature(agentcontext):
 
     tl = TechLead(sm, ui)
     response = await tl.run()
-    assert response.type == ResponseType.DONE
+    assert response.type == ResponseType.UPDATE_SPECIFICATION
 
     await sm.commit()
 

tests/templates/test_templates.py

@@ -6,6 +6,7 @@ from core.state.state_manager import StateManager
 from core.templates.registry import PROJECT_TEMPLATES
 
 
+@pytest.mark.skip
 @pytest.mark.asyncio
 @patch("core.state.state_manager.get_config")
 async def test_render_react_express_sql(mock_get_config, testmanager):
@@ -30,6 +31,7 @@ async def test_render_react_express_sql(mock_get_config, testmanager):
     assert "api/models/user.js" not in files
 
 
+@pytest.mark.skip
 @pytest.mark.asyncio
 @patch("core.state.state_manager.get_config")
 async def test_render_react_express_nosql(mock_get_config, testmanager):