Restructre SWE Agent (#204)

### **Description** - Enhanced the `SubmitPatchRequest` and `SubmitPatch` class descriptions to include details about the diff format. - Simplified `CoderAgentArgs` initialization and updated method calls across multiple files. - Introduced `BaseSWEAgent` abstract class. - Added `REVIEWER_BACKSTORY_TMPL` template for reviewer agent. - Refactored `CoderAgent` to inherit from `BaseSWEAgent` and added reviewer agent to the workflow. - Added utility functions for logger setup and LLM retrieval. - Updated README to include additional setup steps and environment variables. - Changed the patch generated to be static instead of agentic. ___ ### **Changes walkthrough** 📝 <table><thead><tr><th></th><th align="left">Relevant files</th></tr></thead><tbody><tr><td><strong>Enhancement </strong></td><td><details><summary>8 files</summary><table> <tr> <td> <details> <summary><strong>submit_patch.py</strong><dd><code>Improve patch submission descriptions and examples</code>              </dd></summary> <hr> python/composio/local_tools/local_workspace/submit_patch/actions/submit_patch.py <li>Enhanced the <code>SubmitPatchRequest</code> and <code>SubmitPatch</code> class descriptions to <br>include details about the diff format.<br> </details> </td> <td><a href="https://github.com/ComposioHQ/composio/pull/204/files#diff-eb4ff8e8a15704392fa1a2d3795544db0b1f85b04299099ca1d67bb0395a4d43">+11/-2</a>    </td> </tr> <tr> <td> <details> <summary><strong>run_evaluation.py</strong><dd><code>Simplify CoderAgentArgs initialization and method call</code>      </dd></summary> <hr> python/composio_swe/benchmark/run_evaluation.py <li>Simplified <code>CoderAgentArgs</code> initialization by removing redundant <br>parameters.<br> <li> Updated <code>coder.run</code> method to accept <code>issue_config</code>.<br> </details> </td> <td><a href="https://github.com/ComposioHQ/composio/pull/204/files#diff-d97c3a8fcc8931e4bc7bd8a5fd22231faf0ffd60133f65afb29ce91ee372ca87">+2/-6</a>      </td> </tr> <tr> <td> <details> <summary><strong>base_swe_agent.py</strong><dd><code>Add BaseSWEAgent abstract class</code>                                                    </dd></summary> <hr> python/composio_swe/composio_swe/agent/base_swe_agent.py <li>Introduced <code>BaseSWEAgent</code> abstract class with an abstract <code>run</code> method.<br> </details> </td> <td><a href="https://github.com/ComposioHQ/composio/pull/204/files#diff-8b58d3656c4f5dce0d84109cfd9952e5d55c3c3075c4bb18bd26e2840896100b">+12/-0</a>    </td> </tr> <tr> <td> <details> <summary><strong>prompts.py</strong><dd><code>Add reviewer backstory template</code>                                                    </dd></summary> <hr> python/composio_swe/composio_swe/agent/prompts.py - Added `REVIEWER_BACKSTORY_TMPL` template for reviewer agent. </details> </td> <td><a href="https://github.com/ComposioHQ/composio/pull/204/files#diff-a95507b8fd735f05687d8673e16728d15b6099e5f5a3ec9313f1afb3d3ec0a27">+12/-0</a>    </td> </tr> <tr> <td> <details> <summary><strong>swe.py</strong><dd><code>Refactor CoderAgent and add reviewer agent</code>                              </dd></summary> <hr> python/composio_swe/composio_swe/agent/swe.py <li>Refactored <code>CoderAgent</code> to inherit from <code>BaseSWEAgent</code>.<br> <li> Moved logger setup and LLM retrieval to <code>utils.py</code>.<br> <li> Added reviewer agent and task to the workflow.<br> </details> </td> <td><a href="https://github.com/ComposioHQ/composio/pull/204/files#diff-bbb8fcd0bfdb219cf4e20ffd41e3695abb277d9a5efc2c5a315e478976655540">+59/-80</a>  </td> </tr> <tr> <td> <details> <summary><strong>swe_run.py</strong><dd><code>Update CoderAgentArgs initialization and method call</code>          </dd></summary> <hr> python/composio_swe/composio_swe/agent/swe_run.py <li>Updated <code>CoderAgentArgs</code> initialization to remove redundant parameters.<br> <li> Updated <code>coder_agent.run</code> method to accept <code>issue_config</code>.<br> </details> </td> <td><a href="https://github.com/ComposioHQ/composio/pull/204/files#diff-784e85fde02b297e5d9c252831add5cfe880f537076facd0cd3f90ac62121e86">+4/-5</a>      </td> </tr> <tr> <td> <details> <summary><strong>utils.py</strong><dd><code>Add utility functions for logger and LLM</code>                                  </dd></summary> <hr> python/composio_swe/composio_swe/agent/utils.py - Added utility functions for logger setup and LLM retrieval. </details> </td> <td><a href="https://github.com/ComposioHQ/composio/pull/204/files#diff-ebba6012e3781136b15802f8496f063323a2d11aba51c645d1d8a5819486dab2">+34/-0</a>    </td> </tr> <tr> <td> <details> <summary><strong>cli.py</strong><dd><code>Simplify solve command initialization and method call</code>        </dd></summary> <hr> python/composio_swe/composio_swe/cli/cli.py <li>Updated <code>solve</code> command to simplify <code>CoderAgentArgs</code> initialization and <br>method call.<br> </details> </td> <td><a href="https://github.com/ComposioHQ/composio/pull/204/files#diff-9cb2bb57d930a1d094b58d1dba3b28334c6b806db91bb30ad984da874a046322">+5/-6</a>      </td> </tr> </table></details></td></tr><tr><td><strong>Documentation </strong></td><td><details><summary>1 files</summary><table> <tr> <td> <details> <summary><strong>README.md</strong><dd><code>Update README with additional setup steps</code>                                </dd></summary> <hr> python/composio_swe/composio_swe/agent/README.md <li>Updated README to include additional setup steps and environment <br>variables.<br> </details> </td> <td><a href="https://github.com/ComposioHQ/composio/pull/204/files#diff-3ee62cdba3c5db57de103a824f94d873c44d8a03368853305b126b5d5e665ec0">+14/-6</a>    </td> </tr> </table></details></td></tr></tr></tbody></table> ___ > 💡 **PR-Agent usage**: >Comment `/help` on the PR to get a list of all available PR-Agent tools and their descriptions
ComposioHQ · Jun 24, 2024 · 702abbf · 702abbf
1 parent ca96505
commit 702abbf
Show file tree

Hide file tree

Showing 17 changed files with 594 additions and 205 deletions.
diff --git a/python/composio/client/enums.py b/python/composio/client/enums.py
@@ -7221,19 +7221,19 @@ def from_app_and_action(cls, app: str, name: str) -> "Action":
     GMAIL_LIST_LABELS = ("gmail", "gmail_list_labels", False)
     GMAIL_REPLY_TO_THREAD = ("gmail", "gmail_reply_to_thread", False)
     GMAIL_SEND_EMAIL = ("gmail", "gmail_send_email", False)
-    GOOGLECALENDAR_CREATE_GOOGLE_EVENT = (
+    GOOGLECALENDAR_CREATE_EVENT = (
         "googlecalendar",
-        "googlecalendar_create_google_event",
+        "googlecalendar_create_event",
         False,
     )
-    GOOGLECALENDAR_DELETE_GOOGLE_EVENT = (
+    GOOGLECALENDAR_DELETE_EVENT = (
         "googlecalendar",
-        "googlecalendar_delete_google_event",
+        "googlecalendar_delete_event",
         False,
     )
-    GOOGLECALENDAR_DUPLICATE_GOOGLE_CALENDAR = (
+    GOOGLECALENDAR_DUPLICATE_CALENDAR = (
         "googlecalendar",
-        "googlecalendar_duplicate_google_calendar",
+        "googlecalendar_duplicate_calendar",
         False,
     )
     GOOGLECALENDAR_FIND_EVENT = ("googlecalendar", "googlecalendar_find_event", False)
@@ -7247,29 +7247,25 @@ def from_app_and_action(cls, app: str, name: str) -> "Action":
         "googlecalendar_get_current_date_time",
         False,
     )
-    GOOGLECALENDAR_LIST_GOOGLE_CALENDARS = (
+    GOOGLECALENDAR_LIST_CALENDARS = (
         "googlecalendar",
-        "googlecalendar_list_google_calendars",
+        "googlecalendar_list_calendars",
         False,
     )
-    GOOGLECALENDAR_PATCH_GOOGLE_CALENDAR = (
+    GOOGLECALENDAR_PATCH_CALENDAR = (
         "googlecalendar",
-        "googlecalendar_patch_google_calendar",
-        False,
-    )
-    GOOGLECALENDAR_QUICK_ADD_GOOGLE_CALENDAR = (
-        "googlecalendar",
-        "googlecalendar_quick_add_google_calendar",
+        "googlecalendar_patch_calendar",
         False,
     )
+    GOOGLECALENDAR_QUICK_ADD = ("googlecalendar", "googlecalendar_quick_add", False)
     GOOGLECALENDAR_REMOVE_ATTENDEE = (
         "googlecalendar",
         "googlecalendar_remove_attendee",
         False,
     )
-    GOOGLECALENDAR_UPDATE_GOOGLE_EVENT = (
+    GOOGLECALENDAR_UPDATE_EVENT = (
         "googlecalendar",
-        "googlecalendar_update_google_event",
+        "googlecalendar_update_event",
         False,
     )
     GOOGLEDOCS_CREATE_DOCUMENT = ("googledocs", "googledocs_create_document", False)
@@ -13830,12 +13826,6 @@ def from_app_and_action(cls, app: str, name: str) -> "Action":
         True,
         True,
     )
-    CMDMANAGERTOOL_GITREPOTREE = (
-        "cmdmanagertool",
-        "cmdmanagertool_gitrepotree",
-        True,
-        True,
-    )
     CMDMANAGERTOOL_EDITFILE = ("cmdmanagertool", "cmdmanagertool_editfile", True, True)
     CMDMANAGERTOOL_RUNCOMMANDONWORKSPACE = (
         "cmdmanagertool",
@@ -13855,6 +13845,18 @@ def from_app_and_action(cls, app: str, name: str) -> "Action":
         True,
         True,
     )
+    CMDMANAGERTOOL_GITREPOTREE = (
+        "cmdmanagertool",
+        "cmdmanagertool_gitrepotree",
+        True,
+        True,
+    )
+    CMDMANAGERTOOL_GETPATCHCMD = (
+        "cmdmanagertool",
+        "cmdmanagertool_getpatchcmd",
+        True,
+        True,
+    )
     HISTORYKEEPER_GETWORKSPACEHISTORY = (
         "historykeeper",
         "historykeeper_getworkspacehistory",
@@ -13904,6 +13906,8 @@ def event(self) -> str:
     GITHUB_ISSUE_ADDED_EVENT = ("github", "github_issue_added_event")
     GITHUB_LABEL_ADDED_EVENT = ("github", "github_label_added_event")
     GITHUB_PULL_REQUEST_EVENT = ("github", "github_pull_request_event")
+    GITHUB_STAR_ADDED_EVENT = ("github", "github_star_added_event")
+    GMAIL_NEW_GMAIL_MESSAGE = ("gmail", "gmail_new_gmail_message")
     GOOGLEDRIVE_GOOGLE_DRIVE_CHANGES = (
         "googledrive",
         "googledrive_google_drive_changes",

diff --git a/python/composio/local_tools/local_workspace/cmd_manager/actions/__init__.py b/python/composio/local_tools/local_workspace/cmd_manager/actions/__init__.py
@@ -1,6 +1,7 @@
 from .clone_github import GithubCloneCmd
 from .cmds import CreateFileCmd, GoToLineNumInOpenFile, OpenFile
 from .edit_cmd import EditFile
+from .get_patch import GetPatchCmd
 from .git_tree import GitRepoTree
 from .run_cmd import RunCommandOnWorkspace
 from .scroll_cmds import Scroll

diff --git a/python/composio/local_tools/local_workspace/cmd_manager/actions/clone_github.py b/python/composio/local_tools/local_workspace/cmd_manager/actions/clone_github.py
@@ -31,6 +31,10 @@ class GithubCloneRequest(BaseRequest):
         description="after cloning the git repo, repo will be set to this commit-id."
         "if commit-id is empty, default branch of the repo will be cloned",
     )
+    just_reset: bool = Field(
+        False,
+        description="If true, the repo will not be cloned. It will be assumed to exist. The repo will be cleaned and reset to the given commit-id",
+    )
 
 
 class GithubCloneResponse(BaseResponse):
@@ -51,6 +55,11 @@ class GithubCloneCmd(BaseAction):
     def execute(
         self, request_data: GithubCloneRequest, authorisation_data: dict
     ) -> BaseResponse:
+        if request_data.just_reset:
+            print("Resetting repository to base commit")
+            self.reset_to_base_commit(request_data)
+            return BaseResponse(output="Repository reset to base commit", return_code=0)
+
         if not request_data.repo_name or not request_data.repo_name.strip():
             raise ValueError("repo_name can not be null. Give a repo_name to clone")
 
@@ -81,3 +90,32 @@ def execute(
         )
         output, return_code = process_output(output, return_code)
         return BaseResponse(output=output, return_code=return_code)
+
+    def reset_to_base_commit(self, request_data: GithubCloneRequest) -> None:
+        """
+        Resets the repository to the specified base commit and cleans any untracked files or changes.
+        Assumes the repository already exists as cloned by the execute function.
+        """
+        print("Resetting repository to base commit inside reset_to_base_commit")
+        self._setup(request_data)
+        # repo_dir = request_data.repo_name.split("/")[-1].strip()
+        reset_commands = [
+            "git fetch --all",
+            f"git reset --hard {request_data.commit_id}",
+            "git clean -fdx",
+        ]
+        if self.container_process is None:
+            raise ValueError("Container process is not set")
+        print("Resetting repository to base commit checked container process")
+        reset_command = " && ".join(reset_commands)
+        output, return_code = communicate(
+            self.container_process,
+            self.container_obj,
+            reset_command,
+            self.parent_pids,
+            timeout_duration=LONG_TIMEOUT,
+        )
+        print(f"Resetting repository to base commit output: {output}")
+        if return_code != 0:
+            raise RuntimeError(f"Failed to reset repository: {output}")
+        print("Repository successfully reset to base commit and cleaned.")
diff --git a/python/composio/local_tools/local_workspace/cmd_manager/actions/get_patch.py b/python/composio/local_tools/local_workspace/cmd_manager/actions/get_patch.py
@@ -0,0 +1,80 @@
+import typing as t
+
+from pydantic import Field
+
+from composio.local_tools.local_workspace.commons.get_logger import get_logger
+from composio.local_tools.local_workspace.commons.history_processor import (
+    history_recorder,
+)
+from composio.local_tools.local_workspace.commons.local_docker_workspace import (
+    communicate,
+)
+from composio.local_tools.local_workspace.commons.utils import process_output
+
+from .base_class import BaseAction, BaseRequest, BaseResponse
+
+
+LONG_TIMEOUT = 200
+logger = get_logger()
+
+
+class GetPatchRequest(BaseRequest):
+    new_file_path: t.List[str] = Field(
+        default=[],
+        description="Paths of the files newly created to be included in the patch.",
+    )
+
+
+class GetPatchResponse(BaseResponse):
+    pass
+
+
+class GetPatchCmd(BaseAction):
+    """
+    Get the patch from the current working directory. The patch is present in the output field of the response.
+    The patch is in the format of a proper diff format.
+    It incorporates any new files specified in the request, thereby excluding irrelevant installation files.
+    It includes deleted files by default.
+    You should run it after
+    Example:
+    diff --git a/repo/example.py b/repo/example.py
+    index 1234567..89abcde 100644
+    --- a/repo/example.py
+    +++ b/repo/example.py
+    @@ -1 +1 @@
+    -Hello, World!
+    +Hello, Composio!
+    """
+
+    _history_maintains: bool = True
+    _display_name = "Get Patch Action"
+    _request_schema = GetPatchRequest
+    _response_schema = GetPatchResponse
+
+    @history_recorder()
+    def execute(
+        self, request_data: GetPatchRequest, authorisation_data: dict
+    ) -> BaseResponse:
+        print("Get patch command...")
+        self._setup(request_data)
+        print("Setup completed.")
+        if self.container_process is None:
+            raise ValueError("Container process is not set")
+        new_files = " ".join(request_data.new_file_path)
+        cmd1 = "git add -u"
+        if len(request_data.new_file_path) > 0:
+            cmd1 = f"git add {new_files} && " + cmd1
+        output, return_code = communicate(
+            self.container_process, self.container_obj, cmd1, self.parent_pids
+        )
+        output, return_code = process_output(output, return_code)
+        print(f"Output of git add: {output}")
+        cmd2 = "git diff --cached"
+        output, return_code = communicate(
+            self.container_process, self.container_obj, cmd2, self.parent_pids
+        )
+        output, return_code = process_output(output, return_code)
+        return BaseResponse(
+            output=output,
+            return_code=return_code,
+        )
diff --git a/python/composio/local_tools/local_workspace/cmd_manager/actions/git_tree.py b/python/composio/local_tools/local_workspace/cmd_manager/actions/git_tree.py
@@ -41,8 +41,7 @@ def execute(
             self.container_process, self.container_obj, self.command, self.parent_pids
         )
         output, return_code = process_output(output, return_code)
-        files = output.split("\n")
         return BaseResponse(
-            output=files,
+            output="Check git_repo_tree.txt for the git-repo-tree results. Use Open File function to check the file.",
             return_code=return_code,
         )
diff --git a/python/composio/local_tools/local_workspace/cmd_manager/tool.py b/python/composio/local_tools/local_workspace/cmd_manager/tool.py
@@ -6,6 +6,7 @@
     EditFile,
     FindFileCmd,
     GetCurrentDirCmd,
+    GetPatchCmd,
     GitRepoTree,
     GithubCloneCmd,
     GoToLineNumInOpenFile,
@@ -43,6 +44,7 @@ def actions(self) -> list:
             GetCurrentDirCmd,
             GithubCloneCmd,
             GitRepoTree,
+            GetPatchCmd,
         ]
 
     def triggers(self) -> list:

diff --git a/python/composio/local_tools/local_workspace/submit_patch/actions/submit_patch.py b/python/composio/local_tools/local_workspace/submit_patch/actions/submit_patch.py
@@ -19,7 +19,10 @@ class SubmitPatchRequest(BaseModel):
     workspace_id: str = Field(
         ..., description="workspace-id for which patch is generated"
     )
-    patch_code: str = Field(..., description="patch code that needs to be submitted")
+    patch_code: str = Field(
+        ...,
+        description="patch code that needs to be submitted - this should be a valid patch in diff format for the workspace.",
+    )
 
 
 class SubmitPatchResponse(BaseModel):
@@ -28,7 +31,13 @@ class SubmitPatchResponse(BaseModel):
 
 class SubmitPatch(Action):
     """
-    Submits generated patch for the workspace that should work
+    Submits generated patch for the workspace that should work This should be in diff format.
+    Example:
+    --- a/file.txt
+    +++ b/file.txt
+    @@ -1,3 +1,3 @@
+    -Hello, World!
+    +Hello, Python!
     """
 
     _history_maintains = True