Add actions for github push and send PR (#1415)

* Added a push action

* Tests

* Add tests

* Fix capitalization

* Update

* Fix typo

* Fix integration tests

* Added poetry.lock

* Set lock

* Fix action parsing

* Update integration test output

* Updated prompt

* Update integration test

* Add github token to default config

---------

Co-authored-by: Engel Nyst <[email protected]>

agenthub/dummy_agent/agent.py

@@ -0,0 +1,21 @@
+"""Module for a Dummy agent."""
+
+from opendevin.action.base import NullAction
+from opendevin.state import State
+from opendevin.action import Action
+from typing import List
+from opendevin.agent import Agent
+from opendevin.controller.agent_controller import AgentController
+from opendevin.observation.base import NullObservation, Observation
+
+class DummyAgent(Agent):
+ """A dummy agent that does nothing but can be used in testing."""
+
+ async def run(self, controller: AgentController) -> Observation:
+ return NullObservation('')
+
+ def step(self, state: State) -> Action:
+ return NullAction('')
+
+ def search_memory(self, query: str) -> List[str]:
+ return []

agenthub/monologue_agent/agent.py

@@ -2,7 +2,7 @@
 from opendevin.agent import Agent
 from opendevin.state import State
 from opendevin.llm.llm import LLM
-from opendevin.schema import ActionType, ObservationType
+from opendevin.schema import ActionType
 from opendevin.exceptions import AgentNoInstructionError
 from opendevin.schema.config import ConfigType
 from opendevin import config
@@ -15,6 +15,7 @@
  FileReadAction,
  AgentRecallAction,
  BrowseURLAction,
+ GitHubPushAction,
  AgentThinkAction,
 )
 
@@ -71,6 +72,10 @@
  'BROWSE google.com',
  '<form><input type="text"></input><button type="submit"></button></form>',
  'I can browse the web too!',
+ 'If I have done some work and I want to push it to github, I can do that also!',
+ "Let's do it.",
+ 'PUSH owner/repo branch',
+ 'The repo was successfully pushed to https://github.com/owner/repo/branch',
  'And once I have completed my task, I can use the finish action to stop working.',
  "But I should only use the finish action when I'm absolutely certain that I've completed my task and have tested my work.",
  'Very cool. Now to accomplish my task.',
@@ -152,32 +157,32 @@ def _initialize(self, task: str):
  else:
  self.memory = None
 
- output_type = ''
+ previous_action = ''
  for thought in INITIAL_THOUGHTS:
  thought = thought.replace('$TASK', task)
- if output_type != '':
+ if previous_action != '':
  observation: Observation = NullObservation(content='')
- if output_type == ObservationType.RUN:
+ if previous_action in {ActionType.RUN, ActionType.PUSH}:
  observation = CmdOutputObservation(
  content=thought, command_id=0, command=''
  )
- elif output_type == ObservationType.READ:
+ elif previous_action == ActionType.READ:
  observation = FileReadObservation(content=thought, path='')
- elif output_type == ObservationType.RECALL:
+ elif previous_action == ActionType.RECALL:
  observation = AgentRecallObservation(
  content=thought, memories=[])
- elif output_type == ObservationType.BROWSE:
+ elif previous_action == ActionType.BROWSE:
  observation = BrowserOutputObservation(
  content=thought, url='', screenshot=''
  )
  self._add_event(observation.to_memory())
- output_type = ''
+ previous_action = ''
  else:
  action: Action = NullAction()
  if thought.startswith('RUN'):
  command = thought.split('RUN ')[1]
  action = CmdRunAction(command)
- output_type = ActionType.RUN
+ previous_action = ActionType.RUN
  elif thought.startswith('WRITE'):
  parts = thought.split('WRITE ')[1].split(' > ')
  path = parts[1]
@@ -186,15 +191,20 @@ def _initialize(self, task: str):
  elif thought.startswith('READ'):
  path = thought.split('READ ')[1]
  action = FileReadAction(path=path)
- output_type = ActionType.READ
+ previous_action = ActionType.READ
  elif thought.startswith('RECALL'):
  query = thought.split('RECALL ')[1]
  action = AgentRecallAction(query=query)
- output_type = ActionType.RECALL
+ previous_action = ActionType.RECALL
  elif thought.startswith('BROWSE'):
  url = thought.split('BROWSE ')[1]
  action = BrowseURLAction(url=url)
- output_type = ActionType.BROWSE
+ previous_action = ActionType.BROWSE
+ elif thought.startswith('PUSH'):
+ owner_repo, branch = thought.split('PUSH ')[1].split(' ')
+ owner, repo = owner_repo.split('/')
+ action = GitHubPushAction(owner=owner, repo=repo, branch=branch)
+ previous_action = ActionType.PUSH
  else:
  action = AgentThinkAction(thought=thought)
  self._add_event(action.to_memory())

agenthub/monologue_agent/utils/prompts.py

@@ -46,6 +46,10 @@
  * `id` - the ID of the background command to kill
 * `browse` - opens a web page. Arguments:
  * `url` - the URL to open
+* `push` - Push a branch from the current repo to github:
+ * `owner` - the owner of the repo to push to
+ * `repo` - the name of the repo to push to
+ * `branch` - the name of the branch to push
 * `recall` - recalls a past memory. Arguments:
  * `query` - the query to search for
 * `think` - make a plan, set a goal, or record your thoughts. Arguments:
@@ -54,7 +58,7 @@
 
 %(background_commands)s
 
-You MUST take time to think in between read, write, run, browse, and recall actions.
+You MUST take time to think in between read, write, run, browse, push, and recall actions.
 You should never act twice in a row without thinking. But if your last several
 actions are all "think" actions, you should consider taking a different action.
 

docs/Agents.md

@@ -15,6 +15,7 @@ Short term memory is stored as a Monologue object and the model can condense it
 `FileReadAction`,
 `AgentRecallAction`,
 `BrowseURLAction`,
+`GithubPushAction`,
 `AgentThinkAction`
 
 ### Observations:
@@ -48,6 +49,7 @@ The agent is given its previous action-observation pairs, current task, and hint
 `CmdRunAction`,
 `CmdKillAction`,
 `BrowseURLAction`,
+`GithubPushAction`,
 `FileReadAction`,
 `FileWriteAction`,
 `AgentRecallAction`,

opendevin/action/__init__.py

@@ -2,6 +2,7 @@
 from .bash import CmdRunAction, CmdKillAction
 from .browse import BrowseURLAction
 from .fileop import FileReadAction, FileWriteAction
+from .github import GitHubPushAction
 from .agent import (
  AgentRecallAction,
  AgentThinkAction,
@@ -25,6 +26,7 @@
  AgentDelegateAction,
  AddTaskAction,
  ModifyTaskAction,
+ GitHubPushAction,
 )
 
 ACTION_TYPE_TO_CLASS = {action_class.action: action_class for action_class in actions} # type: ignore[attr-defined]

opendevin/action/github.py

@@ -0,0 +1,158 @@
+from dataclasses import dataclass
+from opendevin.observation import Observation, AgentErrorObservation
+from opendevin.observation.message import AgentMessageObservation
+from opendevin.observation.run import CmdOutputObservation
+from opendevin.schema import ActionType
+from opendevin import config
+from typing import TYPE_CHECKING
+import requests
+import random
+import string
+
+from opendevin.schema.config import ConfigType
+
+from .base import ExecutableAction
+
+if TYPE_CHECKING:
+ from opendevin.controller import AgentController
+
+
+@dataclass
+class GitHubPushAction(ExecutableAction):
+ """This pushes the current branch to github.
+
+ To use this, you need to set the GITHUB_TOKEN environment variable.
+ The agent will return a message with a URL that you can click to make a pull
+ request.
+
+ Attributes:
+ owner: The owner of the source repo
+ repo: The name of the source repo
+ branch: The branch to push
+ action: The action identifier
+ """
+
+ owner: str
+ repo: str
+ branch: str
+ action: str = ActionType.PUSH
+
+ async def run(self, controller: 'AgentController') -> Observation:
+ github_token = config.get(ConfigType.GITHUB_TOKEN)
+ if not github_token:
+ return AgentErrorObservation(
+ 'GITHUB_TOKEN is not set'
+ )
+
+ # Create a random short string to use as a temporary remote
+ random_remote = ''.join(
+ ['opendevin_temp_'] + random.choices(string.ascii_lowercase, k=5)
+ )
+
+ # Set the temporary remote
+ new_url = f'https://{github_token}@github.com/{self.owner}/{self.repo}.git'
+ command = f'git remote add {random_remote} {new_url}'
+ remote_add_result = controller.action_manager.run_command(
+ command, background=False
+ )
+ if (
+ not isinstance(remote_add_result, CmdOutputObservation)
+ or remote_add_result.exit_code != 0
+ ):
+ return remote_add_result
+
+ # Push the branch to the temporary remote
+ command = f'git push {random_remote} {self.branch}'
+ push_result = controller.action_manager.run_command(command, background=False)
+
+ # Delete the temporary remote
+ command = f'git remote remove {random_remote}'
+ remote_remove_result = controller.action_manager.run_command(
+ command, background=False
+ )
+ if (
+ not isinstance(remote_remove_result, CmdOutputObservation)
+ or remote_remove_result.exit_code != 0
+ ):
+ return remote_remove_result
+
+ return push_result
+
+ @property
+ def message(self) -> str:
+ return f'Pushing branch {self.branch} to {self.owner}/{self.repo}'
+
+
+@dataclass
+class GitHubSendPRAction(ExecutableAction):
+ """An action to send a github PR.
+
+ To use this, you need to set the GITHUB_TOKEN environment variable.
+
+ Attributes:
+ owner: The owner of the source repo
+ repo: The name of the source repo
+ title: The title of the PR
+ head: The branch to send the PR from
+ head_repo: The repo to send the PR from
+ base: The branch to send the PR to
+ body: The body of the PR
+ """
+
+ owner: str
+ repo: str
+ title: str
+ head: str
+ head_repo: str | None
+ base: str
+ body: str | None
+ action: str = ActionType.SEND_PR
+
+ async def run(self, controller: 'AgentController') -> Observation:
+ github_token = config.get(ConfigType.GITHUB_TOKEN)
+ if not github_token:
+ return AgentErrorObservation(
+ 'GITHUB_TOKEN is not set'
+ )
+
+ # API URL to create the pull request
+ url = f'https://api.github.com/repos/{self.owner}/{self.repo}/pulls'
+
+ # Headers to authenticate and request JSON responses
+ headers = {
+ 'Authorization': f'token {github_token}',
+ 'Accept': 'application/vnd.github.v3+json',
+ }
+
+ # Data for the pull request
+ data = {
+ 'title': self.title,
+ 'head': self.head,
+ 'head_repo': self.head_repo,
+ 'base': self.base,
+ 'body': self.body,
+ }
+ data = {k: v for k, v in data.items() if v is not None}
+
+ # Make the request
+ response = requests.post(url, headers=headers, json=data)
+
+ # Check for errors
+ if response.status_code == 201:
+ return AgentMessageObservation(
+ 'Pull request created successfully!\n'
+ f'Pull request URL:{response.json()["html_url"]}'
+ )
+ else:
+ return AgentErrorObservation(
+ 'Failed to create pull request\n'
+ f'Status code: {response.status_code}\n'
+ f'Response: {response.text}'
+ )
+
+ @property
+ def message(self) -> str:
+ return (
+ f'Sending PR from {self.head_repo}:{self.head} to '
+ f'{self.owner}:{self.base}'
+ )

opendevin/config.py

@@ -48,6 +48,7 @@
  ConfigType.USE_HOST_NETWORK: 'false',
  ConfigType.SSH_HOSTNAME: 'localhost',
  ConfigType.DISABLE_COLOR: 'false',
+ ConfigType.GITHUB_TOKEN: None,
 }
 
 config_str = ''

opendevin/schema/action.py

@@ -73,5 +73,11 @@ class ActionTypeSchema(BaseModel):
 
  CHANGE_TASK_STATE: str = Field(default='change_task_state')
 
+ PUSH: str = Field(default='push')
+ """Push a branch to github."""
+
+ SEND_PR: str = Field(default='send_pr')
+ """Send a PR to github."""
+
 
 ActionType = ActionTypeSchema()

opendevin/schema/config.py

@@ -29,3 +29,4 @@ class ConfigType(str, Enum):
  USE_HOST_NETWORK = 'USE_HOST_NETWORK'
  SSH_HOSTNAME = 'SSH_HOSTNAME'
  DISABLE_COLOR = 'DISABLE_COLOR'
+ GITHUB_TOKEN = 'GITHUB_TOKEN'