github/OpenHands
1
0
Fork 0
mirror of https://github.com/OpenHands/OpenHands.git synced 2026-03-22 05:37:20 +08:00
Code Issues Packages Projects Releases Wiki Activity

arch: deprecating recall action and search_memory (#2900)

Browse Source 
* deprecating recall action

* fix integration tests

* fix integration tests

* remove search memory
This commit is contained in:
Xingyao Wang
2024-07-13 03:23:21 +08:00
committed by GitHub
parent 2b7c4e5571
commit e45ddeb2a2
55 changed files with 38 additions and 743 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
agenthub/README.md
View File

@@ -33,7 +33,6 @@ Here is a list of available Actions, which can be returned by `agent.step()`:
- [`FileReadAction`](../opendevin/events/action/files.py) - Reads the content of a file
- [`FileWriteAction`](../opendevin/events/action/files.py) - Writes new content to a file
- [`BrowseURLAction`](../opendevin/events/action/browse.py) - Gets the content of a URL
- [`AgentRecallAction`](../opendevin/events/action/agent.py) - Searches memory (e.g. a vector database)
- [`AddTaskAction`](../opendevin/events/action/tasks.py) - Adds a subtask to the plan
- [`ModifyTaskAction`](../opendevin/events/action/tasks.py) - Changes the state of a subtask.
- [`AgentFinishAction`](../opendevin/events/action/agent.py) - Stops the control loop, allowing the user/delegator agent to enter a new task
@@ -54,7 +53,6 @@ Here is a list of available Observations:
- [`BrowserOutputObservation`](../opendevin/events/observation/browse.py)
- [`FileReadObservation`](../opendevin/events/observation/files.py)
- [`FileWriteObservation`](../opendevin/events/observation/files.py)
- [`AgentRecallObservation`](../opendevin/events/observation/recall.py)
- [`ErrorObservation`](../opendevin/events/observation/error.py)
- [`SuccessObservation`](../opendevin/events/observation/success.py)
@@ -72,14 +70,3 @@ def step(self, state: "State") -> "Action"
`step` moves the agent forward one step towards its goal. This probably means
sending a prompt to the LLM, then parsing the response into an `Action`.
### `search_memory`
```
def search_memory(self, query: str) -> list[str]:
```
`search_memory` should return a list of events that match the query. This will be used
for the `recall` action.
You can optionally just return `[]` for this method, meaning the agent has no long-term memory.

3
agenthub/browsing_agent/browsing_agent.py
View File

@@ -213,6 +213,3 @@ class BrowsingAgent(Agent):
stop=[')```', ')\n```'],
)
return self.response_parser.parse(response)
def search_memory(self, query: str) -> list[str]:
raise NotImplementedError('Implement this abstract method')

3
agenthub/codeact_agent/codeact_agent.py
View File

@@ -208,9 +208,6 @@ class CodeActAgent(Agent):
)
return self.action_parser.parse(response)
def search_memory(self, query: str) -> list[str]:
raise NotImplementedError('Implement this abstract method')
def _get_messages(self, state: State) -> list[dict[str, str]]:
messages = [
{'role': 'system', 'content': self.system_message},

3
agenthub/codeact_swe_agent/codeact_swe_agent.py
View File

@@ -162,9 +162,6 @@ class CodeActSWEAgent(Agent):
return self.response_parser.parse(response)
def search_memory(self, query: str) -> list[str]:
raise NotImplementedError('Implement this abstract method')
def _get_messages(self, state: State) -> list[dict[str, str]]:
messages = [
{'role': 'system', 'content': self.system_message},

3
agenthub/delegator_agent/agent.py
View File

@@ -82,6 +82,3 @@ class DelegatorAgent(Agent):
)
else:
raise Exception('Invalid delegate state')
def search_memory(self, query: str) -> list[str]:
return []

11
agenthub/dummy_agent/agent.py
View File

@@ -7,7 +7,6 @@ from opendevin.events.action import (
Action,
AddTaskAction,
AgentFinishAction,
AgentRecallAction,
AgentRejectAction,
BrowseInteractiveAction,
BrowseURLAction,
@@ -18,7 +17,6 @@ from opendevin.events.action import (
ModifyTaskAction,
)
from opendevin.events.observation import (
AgentRecallObservation,
CmdOutputObservation,
FileReadObservation,
FileWriteObservation,
@@ -91,12 +89,6 @@ class DummyAgent(Agent):
)
],
},
{
'action': AgentRecallAction(query='who am I?'),
'observations': [
AgentRecallObservation('', memories=['I am a computer.']),
],
},
{
'action': BrowseURLAction(url='https://google.com'),
'observations': [
@@ -152,6 +144,3 @@ class DummyAgent(Agent):
hist_obs == expected_obs
), f'Expected observation {expected_obs}, got {hist_obs}'
return self.steps[state.iteration]['action']
def search_memory(self, query: str) -> list[str]:
return ['I am a computer.']

3
agenthub/micro/agent.py
View File

@@ -79,6 +79,3 @@ class MicroAgent(Agent):
action_resp = resp['choices'][0]['message']['content']
action = parse_response(action_resp)
return action
def search_memory(self, query: str) -> list[str]:
return []

23
agenthub/monologue_agent/agent.py
View File

@@ -8,7 +8,6 @@ from opendevin.core.exceptions import AgentNoInstructionError
from opendevin.core.schema import ActionType
from opendevin.events.action import (
Action,
AgentRecallAction,
BrowseURLAction,
CmdRunAction,
FileReadAction,
@@ -17,7 +16,6 @@ from opendevin.events.action import (
NullAction,
)
from opendevin.events.observation import (
AgentRecallObservation,
BrowserOutputObservation,
CmdOutputObservation,
FileReadObservation,
@@ -103,8 +101,6 @@ class MonologueAgent(Agent):
)
elif previous_action == ActionType.READ:
observation = FileReadObservation(content=thought, path='')
elif previous_action == ActionType.RECALL:
observation = AgentRecallObservation(content=thought, memories=[])
elif previous_action == ActionType.BROWSE:
observation = BrowserOutputObservation(
content=thought, url='', screenshot=''
@@ -128,10 +124,6 @@ class MonologueAgent(Agent):
path = thought.split('READ ')[1]
action = FileReadAction(path=path)
previous_action = ActionType.READ
elif thought.startswith('RECALL'):
query = thought.split('RECALL ')[1]
action = AgentRecallAction(query=query)
previous_action = ActionType.RECALL
elif thought.startswith('BROWSE'):
url = thought.split('BROWSE ')[1]
action = BrowseURLAction(url=url)
@@ -192,21 +184,6 @@ class MonologueAgent(Agent):
self.latest_action = action
return action
def search_memory(self, query: str) -> list[str]:
"""
Uses VectorIndexRetriever to find related memories within the long term memory.
Uses search to produce top 10 results.
Parameters:
- The query that we want to find related memories for
Returns:
- A list of top 10 text results that matched the query
"""
if self.memory is None:
return []
return self.memory.search(query)
def reset(self) -> None:
super().reset()

14
agenthub/monologue_agent/utils/prompts.py
View File

@@ -35,14 +35,12 @@ Here are the possible actions:
* `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
* `message` - make a plan, set a goal, record your thoughts, or ask for more input from the user. Arguments:
* `content` - the message to record
* `wait_for_response` - set to `true` to wait for the user to respond before proceeding
* `finish` - if you're absolutely certain that you've completed your task and have tested your work, use the finish action to stop working.
You MUST take time to think in between read, write, run, browse, push, and recall actions--do this with the `message` action.
You MUST take time to think in between read, write, run, browse, and push actions--do this with the `message` action.
You should never act twice in a row without thinking. But if your last several
actions are all `message` actions, you should consider taking a different action.
@@ -92,15 +90,7 @@ INITIAL_THOUGHTS = [
'It seems like I have some kind of short term memory.',
'Each of my thoughts seems to be stored in a JSON array.',
'It seems whatever I say next will be added as an object to the list.',
'But no one has perfect short-term memory. My list of thoughts will be summarized and condensed over time, losing information in the process.',
'Fortunately I have long term memory!',
'I can just perform a recall action, followed by the thing I want to remember. And then related thoughts just spill out!',
"Sometimes they're random thoughts that don't really have to do with what I wanted to remember. But usually they're exactly what I need!",
"Let's try it out!",
'RECALL what it is I want to do',
"Here's what I want to do: $TASK",
'How am I going to get there though?',
"Neat! And it looks like it's easy for me to use the command line too! I just have to perform a run action and include the command I want to run in the command argument. The command output just jumps into my head!",
"It looks like it's easy for me to use the command line too! I just have to perform a run action and include the command I want to run in the command argument. The command output just jumps into my head!",
'RUN echo "hello world"',
'hello world',
'Cool! I bet I can write files too using the write action.',

3
agenthub/planner_agent/agent.py
View File

@@ -49,6 +49,3 @@ class PlannerAgent(Agent):
messages = [{'content': prompt, 'role': 'user'}]
resp = self.llm.completion(messages=messages)
return self.response_parser.parse(resp)
def search_memory(self, query: str) -> list[str]:
return []

3
agenthub/planner_agent/prompt.py
View File

@@ -89,7 +89,7 @@ It must be an object, and it must contain two fields:
* `state` - set to 'in_progress' to start the task, 'completed' to finish it, 'verified' to assert that it was successful, 'abandoned' to give up on it permanently, or `open` to stop working on it for now.
* `finish` - if ALL of your tasks and subtasks have been verified or abandoned, and you're absolutely certain that you've completed your task and have tested your work, use the finish action to stop working.
You MUST take time to think in between read, write, run, browse, and recall actions--do this with the `message` action.
You MUST take time to think in between read, write, run, and browse actions--do this with the `message` action.
You should never act twice in a row without thinking. But if your last several
actions are all `message` actions, you should consider taking a different action.
@@ -109,7 +109,6 @@ def get_hint(latest_action_id: str) -> str:
ActionType.WRITE: 'You just changed a file. You should think about how it affects your plan.',
ActionType.BROWSE: 'You should think about the page you just visited, and what you learned from it.',
ActionType.MESSAGE: "Look at your last thought in the history above. What does it suggest? Don't think anymore--take action.",
ActionType.RECALL: 'You should think about the information you just recalled, and how it should affect your plan.',
ActionType.ADD_TASK: 'You should think about the next action to take.',
ActionType.MODIFY_TASK: 'You should think about the next action to take.',
ActionType.SUMMARIZE: '',

7
docs/i18n/fr/docusaurus-plugin-content-docs/current/usage/agents.md
View File

@@ -55,7 +55,6 @@ _Exemple de CodeActAgent avec `gpt-4-turbo-2024-04-09` effectuant une tâche de
| ---------------- | ----------------------------------------------------------------------------------------------------------------------------------------------- |
| `__init__` | Initialise un agent avec `llm` et une liste de messages `list[Mapping[str, str]]` |
| `step` | Effectue une étape en utilisant l'agent CodeAct. Cela inclut la collecte d'informations sur les étapes précédentes et invite le modèle à exécuter une commande. |
| `search_memory` | Pas encore implémenté |
### En cours de réalisation & prochaine étape
@@ -77,7 +76,6 @@ La mémoire à court terme est stockée en tant qu'objet Monologue et le modèle
`CmdRunAction`,
`FileWriteAction`,
`FileReadAction`,
`AgentRecallAction`,
`BrowseURLAction`,
`GithubPushAction`,
`AgentThinkAction`
@@ -88,7 +86,6 @@ La mémoire à court terme est stockée en tant qu'objet Monologue et le modèle
`NullObservation`,
`CmdOutputObservation`,
`FileReadObservation`,
`AgentRecallObservation`,
`BrowserOutputObservation`
### Méthodes
@@ -99,7 +96,6 @@ La mémoire à court terme est stockée en tant qu'objet Monologue et le modèle
| `_add_event` | Ajoute des événements au monologue de l'agent et condense avec un résumé automatiquement si le monologue est trop long |
| `_initialize` | Utilise la liste `INITIAL_THOUGHTS` pour donner à l'agent un contexte pour ses capacités et comment naviguer dans le `/workspace` |
| `step` | Modifie l'état actuel en ajoutant les actions et observations les plus récentes, puis invite le modèle à réfléchir à la prochaine action à entreprendre. |
| `search_memory` | Utilise `VectorIndexRetriever` pour trouver des souvenirs liés à la mémoire à long terme. |
## Agent Planificateur
@@ -116,7 +112,6 @@ L'agent reçoit ses paires action-observation précédentes, la tâche actuelle,
`GithubPushAction`,
`FileReadAction`,
`FileWriteAction`,
`AgentRecallAction`,
`AgentThinkAction`,
`AgentFinishAction`,
`AgentSummarizeAction`,
@@ -129,7 +124,6 @@ L'agent reçoit ses paires action-observation précédentes, la tâche actuelle,
`NullObservation`,
`CmdOutputObservation`,
`FileReadObservation`,
`AgentRecallObservation`,
`BrowserOutputObservation`
### Méthodes
@@ -138,4 +132,3 @@ L'agent reçoit ses paires action-observation précédentes, la tâche actuelle,
| ---------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `__init__` | Initialise un agent avec `llm` |
| `step` | Vérifie si l'étape actuelle est terminée, retourne `AgentFinishAction` si oui. Sinon, crée une incitation de planification et l'envoie au modèle pour inférence, en ajoutant le résultat comme prochaine action. |
| `search_memory` | Pas encore implémenté |

7
docs/i18n/zh-Hans/docusaurus-plugin-content-docs/current/usage/agents.md
View File

@@ -55,7 +55,6 @@ _CodeActAgent使用`gpt-4-turbo-2024-04-09`执行数据科学任务（线性回
| -------------- | ------------------------------------------------------------------------------------------------------------------------------------------ |
| `__init__` | 使用`llm`和一系列信息`list[Mapping[str, str]]`初始化Agent |
| `step` | 使用CodeAct Agent执行一步操作，包括收集前一步的信息并提示模型执行命令。 |
| `search_memory`| 尚未实现 |
### 进行中的工作 & 下一步
@@ -77,7 +76,6 @@ Monologue Agent利用长短期记忆来完成任务。
`CmdRunAction`,
`FileWriteAction`,
`FileReadAction`,
`AgentRecallAction`,
`BrowseURLAction`,
`GithubPushAction`,
`AgentThinkAction`
@@ -88,7 +86,6 @@ Monologue Agent利用长短期记忆来完成任务。
`NullObservation`,
`CmdOutputObservation`,
`FileReadObservation`,
`AgentRecallObservation`,
`BrowserOutputObservation`
### 方法
@@ -99,7 +96,6 @@ Monologue Agent利用长短期记忆来完成任务。
| `_add_event` | 将事件附加到Agent的独白中，如独白过长自动与摘要一起压缩 |
| `_initialize` | 使用`INITIAL_THOUGHTS`列表为agent提供其能力的上下文以及如何导航`/workspace` |
| `step` | 通过添加最近的动作和观测修改当前状态，然后提示模型考虑其接下来的动作。 |
| `search_memory`| 使用`VectorIndexRetriever`在长期记忆中查找相关记忆。 |
## Planner Agent
@@ -116,7 +112,6 @@ Planner agent利用特殊的提示策略为解决问题创建长期计划。
`GithubPushAction`,
`FileReadAction`,
`FileWriteAction`,
`AgentRecallAction`,
`AgentThinkAction`,
`AgentFinishAction`,
`AgentSummarizeAction`,
@@ -129,7 +124,6 @@ Planner agent利用特殊的提示策略为解决问题创建长期计划。
`NullObservation`,
`CmdOutputObservation`,
`FileReadObservation`,
`AgentRecallObservation`,
`BrowserOutputObservation`
### 方法
@@ -138,4 +132,3 @@ Planner agent利用特殊的提示策略为解决问题创建长期计划。
| -------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `__init__` | 使用`llm`初始化Agent |
| `step` | 检查当前步骤是否完成，如果是则返回`AgentFinishAction`。否则，创建计划提示并发送给模型进行推理，将结果作为下一步动作。 |
| `search_memory`| 尚未实现 |

7
docs/modules/usage/agents.md
View File

@@ -55,7 +55,6 @@ _Example of CodeActAgent with `gpt-4-turbo-2024-04-09` performing a data science
| --------------- | ----------------------------------------------------------------------------------------------------------------------------------------------- |
| `__init__` | Initializes an agent with `llm` and a list of messages `list[Mapping[str, str]]` |
| `step` | Performs one step using the CodeAct Agent. This includes gathering info on previous steps and prompting the model to make a command to execute. |
| `search_memory` | Not yet implemented |
## Monologue Agent
@@ -72,7 +71,6 @@ Short term memory is stored as a Monologue object and the model can condense it
`CmdRunAction`,
`FileWriteAction`,
`FileReadAction`,
`AgentRecallAction`,
`BrowseURLAction`,
`GithubPushAction`,
`AgentThinkAction`
@@ -83,7 +81,6 @@ Short term memory is stored as a Monologue object and the model can condense it
`NullObservation`,
`CmdOutputObservation`,
`FileReadObservation`,
`AgentRecallObservation`,
`BrowserOutputObservation`
### Methods
@@ -94,7 +91,6 @@ Short term memory is stored as a Monologue object and the model can condense it
| `_add_event` | Appends events to the monologue of the agent and condenses with summary automatically if the monologue is too long |
| `_initialize` | Utilizes the `INITIAL_THOUGHTS` list to give the agent a context for its capabilities and how to navigate the `/workspace` |
| `step` | Modifies the current state by adding the most recent actions and observations, then prompts the model to think about its next action to take. |
| `search_memory` | Uses `VectorIndexRetriever` to find related memories within the long term memory. |
## Planner Agent
@@ -111,7 +107,6 @@ The agent is given its previous action-observation pairs, current task, and hint
`GithubPushAction`,
`FileReadAction`,
`FileWriteAction`,
`AgentRecallAction`,
`AgentThinkAction`,
`AgentFinishAction`,
`AgentSummarizeAction`,
@@ -124,7 +119,6 @@ The agent is given its previous action-observation pairs, current task, and hint
`NullObservation`,
`CmdOutputObservation`,
`FileReadObservation`,
`AgentRecallObservation`,
`BrowserOutputObservation`
### Methods
@@ -133,4 +127,3 @@ The agent is given its previous action-observation pairs, current task, and hint
| --------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `__init__` | Initializes an agent with `llm` |
| `step` | Checks to see if current step is completed, returns `AgentFinishAction` if True. Otherwise, creates a plan prompt and sends to model for inference, adding the result as the next action. |
| `search_memory` | Not yet implemented |

11
docs/static/img/backend_architecture.puml vendored
View File

@@ -14,15 +14,6 @@ class opendevin.observation.AgentMessageObservation {
role: str
observation: str
}
class opendevin.action.agent.AgentRecallAction {
query: str
action: str
}
class opendevin.observation.AgentRecallObservation {
memories: List[str]
role: str
observation: str
}
class opendevin.action.agent.AgentSummarizeAction {
summary: str
action: str
@@ -173,8 +164,6 @@ class opendevin.server.session.Session {
opendevin.action.base.ExecutableAction <|-- opendevin.action.agent.AgentEchoAction
opendevin.action.base.NotExecutableAction <|-- opendevin.action.agent.AgentFinishAction
opendevin.observation.Observation <|-- opendevin.observation.AgentMessageObservation
opendevin.action.base.ExecutableAction <|-- opendevin.action.agent.AgentRecallAction
opendevin.observation.Observation <|-- opendevin.observation.AgentRecallObservation
opendevin.action.base.NotExecutableAction <|-- opendevin.action.agent.AgentSummarizeAction
opendevin.action.base.NotExecutableAction <|-- opendevin.action.agent.AgentThinkAction
opendevin.action.base.Action <|-- opendevin.action.base.ExecutableAction

3
frontend/src/types/ActionType.tsx
View File

@@ -26,9 +26,6 @@ enum ActionType {
// Delegate a (sub)task to another agent.
DELEGATE = "delegate",
// Searches long-term memory.
RECALL = "recall",
// If you're absolutely certain that you've completed your task and have tested your work,
// use the finish action to stop working.
FINISH = "finish",

3
frontend/src/types/ObservationType.tsx
View File

@@ -11,9 +11,6 @@ enum ObservationType {
// The output of an IPython command
RUN_IPYTHON = "run_ipython",
// The result of a search
RECALL = "recall",
// A message from the user
CHAT = "chat",

13
opendevin/controller/agent.py
View File

@@ -51,19 +51,6 @@ class Agent(ABC):
"""
pass
@abstractmethod
def search_memory(self, query: str) -> list[str]:
"""
Searches the agent's memory for information relevant to the given query.
Parameters:
- query (str): The query to search for in the agent's memory.
Returns:
- response (str): The response to the query.
"""
pass
def reset(self) -> None:
"""
Resets the agent's execution status and clears the history. This method can be used

4
opendevin/core/schema/action.py
View File

@@ -40,10 +40,6 @@ class ActionTypeSchema(BaseModel):
"""Interact with the browser instance.
"""
RECALL: str = Field(default='recall')
"""Searches long-term memory
"""
DELEGATE: str = Field(default='delegate')
"""Delegates a task to another agent.
"""

4
opendevin/core/schema/observation.py
View File

@@ -22,10 +22,6 @@ class ObservationTypeSchema(BaseModel):
"""Runs a IPython cell.
"""
RECALL: str = Field(default='recall')
"""The result of a search
"""
CHAT: str = Field(default='chat')
"""A message from the user
"""

2
opendevin/events/action/__init__.py
View File

@@ -2,7 +2,6 @@ from .action import Action, ActionConfirmationStatus
from .agent import (
AgentDelegateAction,
AgentFinishAction,
AgentRecallAction,
AgentRejectAction,
AgentSummarizeAction,
ChangeAgentStateAction,
@@ -22,7 +21,6 @@ __all__ = [
'BrowseInteractiveAction',
'FileReadAction',
'FileWriteAction',
'AgentRecallAction',
'AgentFinishAction',
'AgentRejectAction',
'AgentDelegateAction',

15
opendevin/events/action/agent.py
View File

@@ -1,5 +1,4 @@
from dataclasses import dataclass, field
from typing import ClassVar
from opendevin.core.schema import ActionType
@@ -19,18 +18,6 @@ class ChangeAgentStateAction(Action):
return f'Agent state changed to {self.agent_state}'
@dataclass
class AgentRecallAction(Action):
query: str
thought: str = ''
action: str = ActionType.RECALL
runnable: ClassVar[bool] = True
@property
def message(self) -> str:
return f"Let me dive into my memories to find what you're looking for! Searching for: '{self.query}'. This might take a moment."
@dataclass
class AgentSummarizeAction(Action):
summary: str
@@ -54,7 +41,7 @@ class AgentFinishAction(Action):
@property
def message(self) -> str:
if self.thought != "":
if self.thought != '':
return self.thought
return "All done! What's next on the agenda?"

2
opendevin/events/observation/__init__.py
View File

@@ -6,7 +6,6 @@ from .empty import NullObservation
from .error import ErrorObservation
from .files import FileReadObservation, FileWriteObservation
from .observation import Observation
from .recall import AgentRecallObservation
from .reject import RejectObservation
from .success import SuccessObservation
@@ -18,7 +17,6 @@ __all__ = [
'BrowserOutputObservation',
'FileReadObservation',
'FileWriteObservation',
'AgentRecallObservation',
'ErrorObservation',
'AgentStateChangedObservation',
'AgentDelegateObservation',

20
opendevin/events/observation/recall.py
View File

@@ -1,20 +0,0 @@
from dataclasses import dataclass
from opendevin.core.schema import ObservationType
from .observation import Observation
@dataclass
class AgentRecallObservation(Observation):
"""
This data class represents a list of memories recalled by the agent.
"""
memories: list[str]
role: str = 'assistant'
observation: str = ObservationType.RECALL
@property
def message(self) -> str:
return 'The agent recalled memories.'

2
opendevin/events/serialization/action.py
View File

@@ -3,7 +3,6 @@ from opendevin.events.action.action import Action
from opendevin.events.action.agent import (
AgentDelegateAction,
AgentFinishAction,
AgentRecallAction,
AgentRejectAction,
ChangeAgentStateAction,
)
@@ -25,7 +24,6 @@ actions = (
BrowseInteractiveAction,
FileReadAction,
FileWriteAction,
AgentRecallAction,
AgentFinishAction,
AgentRejectAction,
AgentDelegateAction,

2
opendevin/events/serialization/observation.py
View File

@@ -9,7 +9,6 @@ from opendevin.events.observation.empty import NullObservation
from opendevin.events.observation.error import ErrorObservation
from opendevin.events.observation.files import FileReadObservation, FileWriteObservation
from opendevin.events.observation.observation import Observation
from opendevin.events.observation.recall import AgentRecallObservation
from opendevin.events.observation.reject import RejectObservation
from opendevin.events.observation.success import SuccessObservation
@@ -20,7 +19,6 @@ observations = (
BrowserOutputObservation,
FileReadObservation,
FileWriteObservation,
AgentRecallObservation,
AgentDelegateObservation,
SuccessObservation,
ErrorObservation,

30
opendevin/runtime/client/runtime.py
View File

@@ -11,7 +11,6 @@ from opendevin.core.config import config
from opendevin.core.logger import opendevin_logger as logger
from opendevin.events import EventSource, EventStream, EventStreamSubscriber
from opendevin.events.action import (
AgentRecallAction,
BrowseInteractiveAction,
BrowseURLAction,
CmdRunAction,
@@ -257,9 +256,6 @@ class EventStreamRuntime(Runtime):
async def browse_interactive(self, action: BrowseInteractiveAction) -> Observation:
return await self.run_action(action)
async def recall(self, action: AgentRecallAction) -> Observation:
return await self.run_action(action)
############################################################################
# Keep the same with other runtimes
############################################################################
@@ -273,25 +269,6 @@ class EventStreamRuntime(Runtime):
'This method is not implemented in the runtime client.'
)
# async def read(self, action: FileReadAction) -> Observation:
# working_dir = self.get_working_directory()
# return await read_file(action.path, working_dir, action.start, action.end)
# async def write(self, action: FileWriteAction) -> Observation:
# working_dir = self.get_working_directory()
# return await write_file(
# action.path, working_dir, action.content, action.start, action.end
# )
# async def browse(self, action: BrowseURLAction) -> Observation:
# return await browse(action, self.browser)
# async def browse_interactive(self, action: BrowseInteractiveAction) -> Observation:
# return await browse(action, self.browser)
# async def recall(self, action: AgentRecallAction) -> Observation:
# return NullObservation('')
############################################################################
# Initialization work inside sandbox image
############################################################################
@@ -363,13 +340,6 @@ async def test_event_stream():
await runtime.run_action(action_browse), extra={'msg_type': 'OBSERVATION'}
)
# Test recall
action_recall = AgentRecallAction(query='who am I?')
logger.info(action_recall, extra={'msg_type': 'ACTION'})
logger.info(
await runtime.run_action(action_recall), extra={'msg_type': 'OBSERVATION'}
)
if __name__ == '__main__':
asyncio.run(test_event_stream())

5
opendevin/runtime/runtime.py
View File

@@ -8,7 +8,6 @@ from opendevin.events import EventStream, EventStreamSubscriber
from opendevin.events.action import (
Action,
ActionConfirmationStatus,
AgentRecallAction,
BrowseInteractiveAction,
BrowseURLAction,
CmdRunAction,
@@ -163,7 +162,3 @@ class Runtime:
@abstractmethod
async def browse_interactive(self, action: BrowseInteractiveAction) -> Observation:
pass
@abstractmethod
async def recall(self, action: AgentRecallAction) -> Observation:
pass

5
opendevin/runtime/server/runtime.py
View File

@@ -1,6 +1,5 @@
from opendevin.core.config import config
from opendevin.events.action import (
AgentRecallAction,
BrowseInteractiveAction,
BrowseURLAction,
CmdRunAction,
@@ -12,7 +11,6 @@ from opendevin.events.observation import (
CmdOutputObservation,
ErrorObservation,
IPythonRunCellObservation,
NullObservation,
Observation,
)
from opendevin.events.stream import EventStream
@@ -114,9 +112,6 @@ class ServerRuntime(Runtime):
async def browse_interactive(self, action: BrowseInteractiveAction) -> Observation:
return await browse(action, self.browser)
async def recall(self, action: AgentRecallAction) -> Observation:
return NullObservation('')
def _run_command(self, command: str) -> Observation:
try:
exit_code, output = self.sandbox.execute(command)

4
opendevin/server/README.md
View File

@@ -62,8 +62,6 @@ This list may grow over time.
* `command` - the command to run
* `browse` - opens a web page.
* `url` - the URL to open
* `recall` - searches long-term memory
* `query` - the query to search for
* `think` - Allows the agent to make a plan, set a goal, or record thoughts
* `thought` - the thought to record
* `finish` - agent signals that the task is completed
@@ -87,6 +85,4 @@ This list may grow over time.
* `run` - the output of a command
* `command` - the command run
* `exit_code` - the exit code of the command
* `recall` - the result of a search
* `query` - the query searched for
* `chat` - a message from the user

4
opendevin/server/listen.py
View File

@@ -227,10 +227,6 @@ async def websocket_endpoint(websocket: WebSocket):
```json
{"action": "browse", "args": {"url": "https://arxiv.org/html/2402.01030v2"}}
```
- Search long-term memory:
```json
{"action": "recall", "args": {"query": "past projects"}}
```
- Add a task to the root_task:
```json
{"action": "add_task", "args": {"task": "Implement feature X"}}

4
tests/integration/mock/DelegatorAgent/test_edits/prompt_004.log
View File

@@ -11,14 +11,14 @@ Fix typos in bad.txt. Do not ask me for confirmation at any point.
Here's a summary of the codebase, as it relates to this task:
The codebase contains a single file named 'bad.txt' which has some typos. The content of 'bad.txt' is:
The codebase contains a single file named 'bad.txt' which has some typos. The content of the file is:
This is a stupid typoo.
Really?
No mor typos!
Enjoy!
The engineering team needs to correct the typos in this file.
Engineers need to correct the typos in this file.
## Available Actions

4
tests/integration/mock/DelegatorAgent/test_edits/prompt_005.log
View File

@@ -11,14 +11,14 @@ Fix typos in bad.txt. Do not ask me for confirmation at any point.
Here's a summary of the codebase, as it relates to this task:
The codebase contains a single file named 'bad.txt' which has some typos. The content of 'bad.txt' is:
The codebase contains a single file named 'bad.txt' which has some typos. The content of the file is:
This is a stupid typoo.
Really?
No mor typos!
Enjoy!
The engineering team needs to correct the typos in this file.
Engineers need to correct the typos in this file.
## Available Actions

4
tests/integration/mock/DelegatorAgent/test_edits/prompt_006.log
View File

@@ -11,14 +11,14 @@ Fix typos in bad.txt. Do not ask me for confirmation at any point.
Here's a summary of the codebase, as it relates to this task:
The codebase contains a single file named 'bad.txt' which has some typos. The content of 'bad.txt' is:
The codebase contains a single file named 'bad.txt' which has some typos. The content of the file is:
This is a stupid typoo.
Really?
No mor typos!
Enjoy!
The engineering team needs to correct the typos in this file.
Engineers need to correct the typos in this file.
## Available Actions

2
tests/integration/mock/DelegatorAgent/test_edits/response_003.log
View File

@@ -2,7 +2,7 @@
"action": "finish",
"args": {
"outputs": {
"summary": "The codebase contains a single file named 'bad.txt' which has some typos. The content of 'bad.txt' is:\n\nThis is a stupid typoo.\nReally?\nNo mor typos!\nEnjoy!\n\nThe engineering team needs to correct the typos in this file."
"summary": "The codebase contains a single file named 'bad.txt' which has some typos. The content of the file is:\n\nThis is a stupid typoo.\nReally?\nNo mor typos!\nEnjoy!\n\nEngineers need to correct the typos in this file."
}
}
}

63
tests/integration/mock/MonologueAgent/test_write_simple_script/prompt_001.log
View File

@@ -70,64 +70,7 @@ This is your internal monologue, in JSON format:
{
"action": "message",
"args": {
"content": "But no one has perfect short-term memory. My list of thoughts will be summarized and condensed over time, losing information in the process.",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Fortunately I have long term memory!",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "I can just perform a recall action, followed by the thing I want to remember. And then related thoughts just spill out!",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Sometimes they're random thoughts that don't really have to do with what I wanted to remember. But usually they're exactly what I need!",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Let's try it out!",
"wait_for_response": false
}
},
{
"action": "recall",
"args": {
"query": "what it is I want to do",
"thought": ""
}
},
{
"observation": "recall",
"content": "Here's what I want to do: Write a shell script 'hello.sh' that prints 'hello'. Do not ask me for confirmation at any point.",
"extras": {
"memories": [],
"role": "assistant"
}
},
{
"action": "message",
"args": {
"content": "How am I going to get there though?",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Neat! And it looks like it's easy for me to use the command line too! I just have to perform a run action and include the command I want to run in the command argument. The command output just jumps into my head!",
"content": "It looks like it's easy for me to use the command line too! I just have to perform a run action and include the command I want to run in the command argument. The command output just jumps into my head!",
"wait_for_response": false
}
},
@@ -349,14 +292,12 @@ Here are the possible actions:
* `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
* `message` - make a plan, set a goal, record your thoughts, or ask for more input from the user. Arguments:
* `content` - the message to record
* `wait_for_response` - set to `true` to wait for the user to respond before proceeding
* `finish` - if you're absolutely certain that you've completed your task and have tested your work, use the finish action to stop working.
You MUST take time to think in between read, write, run, browse, push, and recall actions--do this with the `message` action.
You MUST take time to think in between read, write, run, browse, and push actions--do this with the `message` action.
You should never act twice in a row without thinking. But if your last several
actions are all `message` actions, you should consider taking a different action.

63
tests/integration/mock/MonologueAgent/test_write_simple_script/prompt_002.log
View File

@@ -70,64 +70,7 @@ This is your internal monologue, in JSON format:
{
"action": "message",
"args": {
"content": "But no one has perfect short-term memory. My list of thoughts will be summarized and condensed over time, losing information in the process.",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Fortunately I have long term memory!",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "I can just perform a recall action, followed by the thing I want to remember. And then related thoughts just spill out!",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Sometimes they're random thoughts that don't really have to do with what I wanted to remember. But usually they're exactly what I need!",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Let's try it out!",
"wait_for_response": false
}
},
{
"action": "recall",
"args": {
"query": "what it is I want to do",
"thought": ""
}
},
{
"observation": "recall",
"content": "Here's what I want to do: Write a shell script 'hello.sh' that prints 'hello'. Do not ask me for confirmation at any point.",
"extras": {
"memories": [],
"role": "assistant"
}
},
{
"action": "message",
"args": {
"content": "How am I going to get there though?",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Neat! And it looks like it's easy for me to use the command line too! I just have to perform a run action and include the command I want to run in the command argument. The command output just jumps into my head!",
"content": "It looks like it's easy for me to use the command line too! I just have to perform a run action and include the command I want to run in the command argument. The command output just jumps into my head!",
"wait_for_response": false
}
},
@@ -368,14 +311,12 @@ Here are the possible actions:
* `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
* `message` - make a plan, set a goal, record your thoughts, or ask for more input from the user. Arguments:
* `content` - the message to record
* `wait_for_response` - set to `true` to wait for the user to respond before proceeding
* `finish` - if you're absolutely certain that you've completed your task and have tested your work, use the finish action to stop working.
You MUST take time to think in between read, write, run, browse, push, and recall actions--do this with the `message` action.
You MUST take time to think in between read, write, run, browse, and push actions--do this with the `message` action.
You should never act twice in a row without thinking. But if your last several
actions are all `message` actions, you should consider taking a different action.

63
tests/integration/mock/MonologueAgent/test_write_simple_script/prompt_003.log
View File

@@ -70,64 +70,7 @@ This is your internal monologue, in JSON format:
{
"action": "message",
"args": {
"content": "But no one has perfect short-term memory. My list of thoughts will be summarized and condensed over time, losing information in the process.",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Fortunately I have long term memory!",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "I can just perform a recall action, followed by the thing I want to remember. And then related thoughts just spill out!",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Sometimes they're random thoughts that don't really have to do with what I wanted to remember. But usually they're exactly what I need!",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Let's try it out!",
"wait_for_response": false
}
},
{
"action": "recall",
"args": {
"query": "what it is I want to do",
"thought": ""
}
},
{
"observation": "recall",
"content": "Here's what I want to do: Write a shell script 'hello.sh' that prints 'hello'. Do not ask me for confirmation at any point.",
"extras": {
"memories": [],
"role": "assistant"
}
},
{
"action": "message",
"args": {
"content": "How am I going to get there though?",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Neat! And it looks like it's easy for me to use the command line too! I just have to perform a run action and include the command I want to run in the command argument. The command output just jumps into my head!",
"content": "It looks like it's easy for me to use the command line too! I just have to perform a run action and include the command I want to run in the command argument. The command output just jumps into my head!",
"wait_for_response": false
}
},
@@ -387,14 +330,12 @@ Here are the possible actions:
* `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
* `message` - make a plan, set a goal, record your thoughts, or ask for more input from the user. Arguments:
* `content` - the message to record
* `wait_for_response` - set to `true` to wait for the user to respond before proceeding
* `finish` - if you're absolutely certain that you've completed your task and have tested your work, use the finish action to stop working.
You MUST take time to think in between read, write, run, browse, push, and recall actions--do this with the `message` action.
You MUST take time to think in between read, write, run, browse, and push actions--do this with the `message` action.
You should never act twice in a row without thinking. But if your last several
actions are all `message` actions, you should consider taking a different action.

63
tests/integration/mock/MonologueAgent/test_write_simple_script/prompt_004.log
View File

@@ -70,64 +70,7 @@ This is your internal monologue, in JSON format:
{
"action": "message",
"args": {
"content": "But no one has perfect short-term memory. My list of thoughts will be summarized and condensed over time, losing information in the process.",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Fortunately I have long term memory!",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "I can just perform a recall action, followed by the thing I want to remember. And then related thoughts just spill out!",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Sometimes they're random thoughts that don't really have to do with what I wanted to remember. But usually they're exactly what I need!",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Let's try it out!",
"wait_for_response": false
}
},
{
"action": "recall",
"args": {
"query": "what it is I want to do",
"thought": ""
}
},
{
"observation": "recall",
"content": "Here's what I want to do: Write a shell script 'hello.sh' that prints 'hello'. Do not ask me for confirmation at any point.",
"extras": {
"memories": [],
"role": "assistant"
}
},
{
"action": "message",
"args": {
"content": "How am I going to get there though?",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Neat! And it looks like it's easy for me to use the command line too! I just have to perform a run action and include the command I want to run in the command argument. The command output just jumps into my head!",
"content": "It looks like it's easy for me to use the command line too! I just have to perform a run action and include the command I want to run in the command argument. The command output just jumps into my head!",
"wait_for_response": false
}
},
@@ -395,14 +338,12 @@ Here are the possible actions:
* `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
* `message` - make a plan, set a goal, record your thoughts, or ask for more input from the user. Arguments:
* `content` - the message to record
* `wait_for_response` - set to `true` to wait for the user to respond before proceeding
* `finish` - if you're absolutely certain that you've completed your task and have tested your work, use the finish action to stop working.
You MUST take time to think in between read, write, run, browse, push, and recall actions--do this with the `message` action.
You MUST take time to think in between read, write, run, browse, and push actions--do this with the `message` action.
You should never act twice in a row without thinking. But if your last several
actions are all `message` actions, you should consider taking a different action.

63
tests/integration/mock/MonologueAgent/test_write_simple_script/prompt_005.log
View File

@@ -70,64 +70,7 @@ This is your internal monologue, in JSON format:
{
"action": "message",
"args": {
"content": "But no one has perfect short-term memory. My list of thoughts will be summarized and condensed over time, losing information in the process.",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Fortunately I have long term memory!",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "I can just perform a recall action, followed by the thing I want to remember. And then related thoughts just spill out!",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Sometimes they're random thoughts that don't really have to do with what I wanted to remember. But usually they're exactly what I need!",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Let's try it out!",
"wait_for_response": false
}
},
{
"action": "recall",
"args": {
"query": "what it is I want to do",
"thought": ""
}
},
{
"observation": "recall",
"content": "Here's what I want to do: Write a shell script 'hello.sh' that prints 'hello'. Do not ask me for confirmation at any point.",
"extras": {
"memories": [],
"role": "assistant"
}
},
{
"action": "message",
"args": {
"content": "How am I going to get there though?",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Neat! And it looks like it's easy for me to use the command line too! I just have to perform a run action and include the command I want to run in the command argument. The command output just jumps into my head!",
"content": "It looks like it's easy for me to use the command line too! I just have to perform a run action and include the command I want to run in the command argument. The command output just jumps into my head!",
"wait_for_response": false
}
},
@@ -414,14 +357,12 @@ Here are the possible actions:
* `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
* `message` - make a plan, set a goal, record your thoughts, or ask for more input from the user. Arguments:
* `content` - the message to record
* `wait_for_response` - set to `true` to wait for the user to respond before proceeding
* `finish` - if you're absolutely certain that you've completed your task and have tested your work, use the finish action to stop working.
You MUST take time to think in between read, write, run, browse, push, and recall actions--do this with the `message` action.
You MUST take time to think in between read, write, run, browse, and push actions--do this with the `message` action.
You should never act twice in a row without thinking. But if your last several
actions are all `message` actions, you should consider taking a different action.

63
tests/integration/mock/MonologueAgent/test_write_simple_script/prompt_006.log
View File

@@ -70,64 +70,7 @@ This is your internal monologue, in JSON format:
{
"action": "message",
"args": {
"content": "But no one has perfect short-term memory. My list of thoughts will be summarized and condensed over time, losing information in the process.",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Fortunately I have long term memory!",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "I can just perform a recall action, followed by the thing I want to remember. And then related thoughts just spill out!",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Sometimes they're random thoughts that don't really have to do with what I wanted to remember. But usually they're exactly what I need!",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Let's try it out!",
"wait_for_response": false
}
},
{
"action": "recall",
"args": {
"query": "what it is I want to do",
"thought": ""
}
},
{
"observation": "recall",
"content": "Here's what I want to do: Write a shell script 'hello.sh' that prints 'hello'. Do not ask me for confirmation at any point.",
"extras": {
"memories": [],
"role": "assistant"
}
},
{
"action": "message",
"args": {
"content": "How am I going to get there though?",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Neat! And it looks like it's easy for me to use the command line too! I just have to perform a run action and include the command I want to run in the command argument. The command output just jumps into my head!",
"content": "It looks like it's easy for me to use the command line too! I just have to perform a run action and include the command I want to run in the command argument. The command output just jumps into my head!",
"wait_for_response": false
}
},
@@ -422,14 +365,12 @@ Here are the possible actions:
* `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
* `message` - make a plan, set a goal, record your thoughts, or ask for more input from the user. Arguments:
* `content` - the message to record
* `wait_for_response` - set to `true` to wait for the user to respond before proceeding
* `finish` - if you're absolutely certain that you've completed your task and have tested your work, use the finish action to stop working.
You MUST take time to think in between read, write, run, browse, push, and recall actions--do this with the `message` action.
You MUST take time to think in between read, write, run, browse, and push actions--do this with the `message` action.
You should never act twice in a row without thinking. But if your last several
actions are all `message` actions, you should consider taking a different action.

63
tests/integration/mock/MonologueAgent/test_write_simple_script/prompt_007.log
View File

@@ -70,64 +70,7 @@ This is your internal monologue, in JSON format:
{
"action": "message",
"args": {
"content": "But no one has perfect short-term memory. My list of thoughts will be summarized and condensed over time, losing information in the process.",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Fortunately I have long term memory!",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "I can just perform a recall action, followed by the thing I want to remember. And then related thoughts just spill out!",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Sometimes they're random thoughts that don't really have to do with what I wanted to remember. But usually they're exactly what I need!",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Let's try it out!",
"wait_for_response": false
}
},
{
"action": "recall",
"args": {
"query": "what it is I want to do",
"thought": ""
}
},
{
"observation": "recall",
"content": "Here's what I want to do: Write a shell script 'hello.sh' that prints 'hello'. Do not ask me for confirmation at any point.",
"extras": {
"memories": [],
"role": "assistant"
}
},
{
"action": "message",
"args": {
"content": "How am I going to get there though?",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Neat! And it looks like it's easy for me to use the command line too! I just have to perform a run action and include the command I want to run in the command argument. The command output just jumps into my head!",
"content": "It looks like it's easy for me to use the command line too! I just have to perform a run action and include the command I want to run in the command argument. The command output just jumps into my head!",
"wait_for_response": false
}
},
@@ -441,14 +384,12 @@ Here are the possible actions:
* `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
* `message` - make a plan, set a goal, record your thoughts, or ask for more input from the user. Arguments:
* `content` - the message to record
* `wait_for_response` - set to `true` to wait for the user to respond before proceeding
* `finish` - if you're absolutely certain that you've completed your task and have tested your work, use the finish action to stop working.
You MUST take time to think in between read, write, run, browse, push, and recall actions--do this with the `message` action.
You MUST take time to think in between read, write, run, browse, and push actions--do this with the `message` action.
You should never act twice in a row without thinking. But if your last several
actions are all `message` actions, you should consider taking a different action.

63
tests/integration/mock/MonologueAgent/test_write_simple_script/prompt_008.log
View File

@@ -70,64 +70,7 @@ This is your internal monologue, in JSON format:
{
"action": "message",
"args": {
"content": "But no one has perfect short-term memory. My list of thoughts will be summarized and condensed over time, losing information in the process.",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Fortunately I have long term memory!",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "I can just perform a recall action, followed by the thing I want to remember. And then related thoughts just spill out!",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Sometimes they're random thoughts that don't really have to do with what I wanted to remember. But usually they're exactly what I need!",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Let's try it out!",
"wait_for_response": false
}
},
{
"action": "recall",
"args": {
"query": "what it is I want to do",
"thought": ""
}
},
{
"observation": "recall",
"content": "Here's what I want to do: Write a shell script 'hello.sh' that prints 'hello'. Do not ask me for confirmation at any point.",
"extras": {
"memories": [],
"role": "assistant"
}
},
{
"action": "message",
"args": {
"content": "How am I going to get there though?",
"wait_for_response": false
}
},
{
"action": "message",
"args": {
"content": "Neat! And it looks like it's easy for me to use the command line too! I just have to perform a run action and include the command I want to run in the command argument. The command output just jumps into my head!",
"content": "It looks like it's easy for me to use the command line too! I just have to perform a run action and include the command I want to run in the command argument. The command output just jumps into my head!",
"wait_for_response": false
}
},
@@ -460,14 +403,12 @@ Here are the possible actions:
* `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
* `message` - make a plan, set a goal, record your thoughts, or ask for more input from the user. Arguments:
* `content` - the message to record
* `wait_for_response` - set to `true` to wait for the user to respond before proceeding
* `finish` - if you're absolutely certain that you've completed your task and have tested your work, use the finish action to stop working.
You MUST take time to think in between read, write, run, browse, push, and recall actions--do this with the `message` action.
You MUST take time to think in between read, write, run, browse, and push actions--do this with the `message` action.
You should never act twice in a row without thinking. But if your last several
actions are all `message` actions, you should consider taking a different action.

2
tests/integration/mock/PlannerAgent/test_write_simple_script/prompt_001.log
View File

@@ -93,7 +93,7 @@ It must be an object, and it must contain two fields:
* `state` - set to 'in_progress' to start the task, 'completed' to finish it, 'verified' to assert that it was successful, 'abandoned' to give up on it permanently, or `open` to stop working on it for now.
* `finish` - if ALL of your tasks and subtasks have been verified or abandoned, and you're absolutely certain that you've completed your task and have tested your work, use the finish action to stop working.
You MUST take time to think in between read, write, run, browse, and recall actions--do this with the `message` action.
You MUST take time to think in between read, write, run, and browse actions--do this with the `message` action.
You should never act twice in a row without thinking. But if your last several
actions are all `message` actions, you should consider taking a different action.

2
tests/integration/mock/PlannerAgent/test_write_simple_script/prompt_002.log
View File

@@ -139,7 +139,7 @@ It must be an object, and it must contain two fields:
* `state` - set to 'in_progress' to start the task, 'completed' to finish it, 'verified' to assert that it was successful, 'abandoned' to give up on it permanently, or `open` to stop working on it for now.
* `finish` - if ALL of your tasks and subtasks have been verified or abandoned, and you're absolutely certain that you've completed your task and have tested your work, use the finish action to stop working.
You MUST take time to think in between read, write, run, browse, and recall actions--do this with the `message` action.
You MUST take time to think in between read, write, run, and browse actions--do this with the `message` action.
You should never act twice in a row without thinking. But if your last several
actions are all `message` actions, you should consider taking a different action.

2
tests/integration/mock/PlannerAgent/test_write_simple_script/prompt_003.log
View File

@@ -150,7 +150,7 @@ It must be an object, and it must contain two fields:
* `state` - set to 'in_progress' to start the task, 'completed' to finish it, 'verified' to assert that it was successful, 'abandoned' to give up on it permanently, or `open` to stop working on it for now.
* `finish` - if ALL of your tasks and subtasks have been verified or abandoned, and you're absolutely certain that you've completed your task and have tested your work, use the finish action to stop working.
You MUST take time to think in between read, write, run, browse, and recall actions--do this with the `message` action.
You MUST take time to think in between read, write, run, and browse actions--do this with the `message` action.
You should never act twice in a row without thinking. But if your last several
actions are all `message` actions, you should consider taking a different action.

2
tests/integration/mock/PlannerAgent/test_write_simple_script/prompt_004.log
View File

@@ -169,7 +169,7 @@ It must be an object, and it must contain two fields:
* `state` - set to 'in_progress' to start the task, 'completed' to finish it, 'verified' to assert that it was successful, 'abandoned' to give up on it permanently, or `open` to stop working on it for now.
* `finish` - if ALL of your tasks and subtasks have been verified or abandoned, and you're absolutely certain that you've completed your task and have tested your work, use the finish action to stop working.
You MUST take time to think in between read, write, run, browse, and recall actions--do this with the `message` action.
You MUST take time to think in between read, write, run, and browse actions--do this with the `message` action.
You should never act twice in a row without thinking. But if your last several
actions are all `message` actions, you should consider taking a different action.

2
tests/integration/mock/PlannerAgent/test_write_simple_script/prompt_005.log
View File

@@ -177,7 +177,7 @@ It must be an object, and it must contain two fields:
* `state` - set to 'in_progress' to start the task, 'completed' to finish it, 'verified' to assert that it was successful, 'abandoned' to give up on it permanently, or `open` to stop working on it for now.
* `finish` - if ALL of your tasks and subtasks have been verified or abandoned, and you're absolutely certain that you've completed your task and have tested your work, use the finish action to stop working.
You MUST take time to think in between read, write, run, browse, and recall actions--do this with the `message` action.
You MUST take time to think in between read, write, run, and browse actions--do this with the `message` action.
You should never act twice in a row without thinking. But if your last several
actions are all `message` actions, you should consider taking a different action.

2
tests/integration/mock/PlannerAgent/test_write_simple_script/prompt_006.log
View File

@@ -187,7 +187,7 @@ It must be an object, and it must contain two fields:
* `state` - set to 'in_progress' to start the task, 'completed' to finish it, 'verified' to assert that it was successful, 'abandoned' to give up on it permanently, or `open` to stop working on it for now.
* `finish` - if ALL of your tasks and subtasks have been verified or abandoned, and you're absolutely certain that you've completed your task and have tested your work, use the finish action to stop working.
You MUST take time to think in between read, write, run, browse, and recall actions--do this with the `message` action.
You MUST take time to think in between read, write, run, and browse actions--do this with the `message` action.
You should never act twice in a row without thinking. But if your last several
actions are all `message` actions, you should consider taking a different action.

2
tests/integration/mock/PlannerAgent/test_write_simple_script/prompt_007.log
View File

@@ -206,7 +206,7 @@ It must be an object, and it must contain two fields:
* `state` - set to 'in_progress' to start the task, 'completed' to finish it, 'verified' to assert that it was successful, 'abandoned' to give up on it permanently, or `open` to stop working on it for now.
* `finish` - if ALL of your tasks and subtasks have been verified or abandoned, and you're absolutely certain that you've completed your task and have tested your work, use the finish action to stop working.
You MUST take time to think in between read, write, run, browse, and recall actions--do this with the `message` action.
You MUST take time to think in between read, write, run, and browse actions--do this with the `message` action.
You should never act twice in a row without thinking. But if your last several
actions are all `message` actions, you should consider taking a different action.

2
tests/integration/mock/PlannerAgent/test_write_simple_script/prompt_008.log
View File

@@ -214,7 +214,7 @@ It must be an object, and it must contain two fields:
* `state` - set to 'in_progress' to start the task, 'completed' to finish it, 'verified' to assert that it was successful, 'abandoned' to give up on it permanently, or `open` to stop working on it for now.
* `finish` - if ALL of your tasks and subtasks have been verified or abandoned, and you're absolutely certain that you've completed your task and have tested your work, use the finish action to stop working.
You MUST take time to think in between read, write, run, browse, and recall actions--do this with the `message` action.
You MUST take time to think in between read, write, run, and browse actions--do this with the `message` action.
You should never act twice in a row without thinking. But if your last several
actions are all `message` actions, you should consider taking a different action.

2
tests/integration/mock/PlannerAgent/test_write_simple_script/prompt_009.log
View File

@@ -224,7 +224,7 @@ It must be an object, and it must contain two fields:
* `state` - set to 'in_progress' to start the task, 'completed' to finish it, 'verified' to assert that it was successful, 'abandoned' to give up on it permanently, or `open` to stop working on it for now.
* `finish` - if ALL of your tasks and subtasks have been verified or abandoned, and you're absolutely certain that you've completed your task and have tested your work, use the finish action to stop working.
You MUST take time to think in between read, write, run, browse, and recall actions--do this with the `message` action.
You MUST take time to think in between read, write, run, and browse actions--do this with the `message` action.
You should never act twice in a row without thinking. But if your last several
actions are all `message` actions, you should consider taking a different action.

2
tests/integration/mock/PlannerAgent/test_write_simple_script/prompt_010.log
View File

@@ -243,7 +243,7 @@ It must be an object, and it must contain two fields:
* `state` - set to 'in_progress' to start the task, 'completed' to finish it, 'verified' to assert that it was successful, 'abandoned' to give up on it permanently, or `open` to stop working on it for now.
* `finish` - if ALL of your tasks and subtasks have been verified or abandoned, and you're absolutely certain that you've completed your task and have tested your work, use the finish action to stop working.
You MUST take time to think in between read, write, run, browse, and recall actions--do this with the `message` action.
You MUST take time to think in between read, write, run, and browse actions--do this with the `message` action.
You should never act twice in a row without thinking. But if your last several
actions are all `message` actions, you should consider taking a different action.

2
tests/integration/mock/PlannerAgent/test_write_simple_script/prompt_011.log
View File

@@ -251,7 +251,7 @@ It must be an object, and it must contain two fields:
* `state` - set to 'in_progress' to start the task, 'completed' to finish it, 'verified' to assert that it was successful, 'abandoned' to give up on it permanently, or `open` to stop working on it for now.
* `finish` - if ALL of your tasks and subtasks have been verified or abandoned, and you're absolutely certain that you've completed your task and have tested your work, use the finish action to stop working.
You MUST take time to think in between read, write, run, browse, and recall actions--do this with the `message` action.
You MUST take time to think in between read, write, run, and browse actions--do this with the `message` action.
You should never act twice in a row without thinking. But if your last several
actions are all `message` actions, you should consider taking a different action.

9
tests/unit/test_action_serialization.py
View File

@@ -3,7 +3,6 @@ from opendevin.events.action import (
Action,
AddTaskAction,
AgentFinishAction,
AgentRecallAction,
AgentRejectAction,
BrowseInteractiveAction,
BrowseURLAction,
@@ -70,14 +69,6 @@ def test_message_action_serialization_deserialization():
serialization_deserialization(original_action_dict, MessageAction)
def test_agent_recall_action_serialization_deserialization():
original_action_dict = {
'action': 'recall',
'args': {'query': 'Test query.', 'thought': ''},
}
serialization_deserialization(original_action_dict, AgentRecallAction)
def test_agent_finish_action_serialization_deserialization():
original_action_dict = {'action': 'finish', 'args': {'outputs': {}, 'thought': ''}}
serialization_deserialization(original_action_dict, AgentFinishAction)