diff --git a/patchwork/common/tools/git_tool.py b/patchwork/common/tools/git_tool.py new file mode 100644 index 000000000..d5acbfa05 --- /dev/null +++ b/patchwork/common/tools/git_tool.py @@ -0,0 +1,49 @@ +from __future__ import annotations + +import os +import subprocess + +from patchwork.common.tools.tool import Tool + + +class GitTool(Tool, tool_name="git_tool", abc_register=False): + def __init__(self, path: str): + super().__init__() + self.path = path + + @property + def json_schema(self) -> dict: + return { + "name": "git_tool", + "description": """\ +Access to the Git CLI, the command is also `git` all args provided are used as is. +""", + "input_schema": { + "type": "object", + "properties": { + "args": { + "type": "array", + "items": {"type": "string"}, + "description": """ +The args to run `git` command with. +E.g. +[\"commit\", \"-m\", \"A commit message\"] to commit changes with a commit message. +[\"add\", \".\"] to stage all changed files. +""", + } + }, + "required": ["args"], + }, + } + + def execute(self, args: list[str]) -> str: + env = os.environ.copy() + p = subprocess.run( + ["git", *args], + env=env, + cwd=self.path, + text=True, + stdout=subprocess.PIPE, + stderr=subprocess.STDOUT, + ) + return p.stdout diff --git a/patchwork/common/tools/github_tool.py b/patchwork/common/tools/github_tool.py index aa5d5effe..16e418c08 100644 --- a/patchwork/common/tools/github_tool.py +++ b/patchwork/common/tools/github_tool.py @@ -6,7 +6,7 @@ from patchwork.common.tools.tool import Tool -class GitHubTool(Tool, tool_name="github_tool"): +class GitHubTool(Tool, tool_name="github_tool", abc_register=False): def __init__(self, path: str, gh_token: str): super().__init__() self.path = path diff --git a/patchwork/steps/DatabaseAgent/README.md b/patchwork/steps/DatabaseAgent/README.md new file mode 100644 index 000000000..f1b4992f0 --- /dev/null +++ b/patchwork/steps/DatabaseAgent/README.md @@ -0,0 +1,56 @@ +# Documentation: DatabaseAgent Module + +## Overview + +The `DatabaseAgent` module is designed to facilitate interaction with a database by generating and executing appropriate SQL queries. It uses machine learning to understand and interpret user tasks and generates an execution plan to gather and process necessary data from a database using the specified database dialect (e.g., SQL). + +The main functionality is encapsulated within a class called `DatabaseAgent`, which leverages the `AgenticStrategyV2` to dynamically generate responses to user tasks. This agent uses the `AioLlmClient` and `DatabaseQueryTool` to perform its operations, geared towards summarizing conversations and fetching data based on user prompts. + +## Inputs + +### DatabaseAgent Inputs Class + +In the file `typed.py`, the `DatabaseAgentInputs` class defines the input structure that the `DatabaseAgent` expects: + +- **task (str):** A string describing the task that needs completion. +- **db_dialect (str):** Specifies the type of SQL dialect the database uses (e.g., MySQL, PostgreSQL). +- **db_driver (str):** The database driver used for connections. +- **db_username (str):** Username for database authentication. +- **db_password (str):** Password for database authentication. +- **db_host (str):** Host address of the database. +- **db_port (int):** Port number for database connection. +- **db_name (str):** Name of the database. +- **db_params (dict):** Additional connection parameters. +- **db_driver_args (dict):** Additional driver-specific arguments. +- **prompt_value (Dict[str, Any]):** Dictionary holding values for prompt customization. +- **max_llm_calls (int):** Configures the maximum number of large language model calls. +- **openai_api_key (str):** API key for OpenAI services. +- **anthropic_api_key (str):** API key for Anthropic services. +- **google_api_key (str):** API key for Google services. +- **example_json (str):** A JSON example to guide response structure. + +## Outputs + +### DatabaseAgent Outputs Class + +Within `typed.py`, the `DatabaseAgentOutputs` class is introduced to outline the expected output format: + +- **request_tokens (int):** Number of tokens used in the request. +- **response_tokens (int):** Number of tokens received in the response. + +## DatabaseAgent Implementation + +### Core Functionality + +In `DatabaseAgent.py`, the `DatabaseAgent` class derives from the `Step` class and uses both `DatabaseAgentInputs` and `DatabaseAgentOutputs`. The class: + +- Initializes an agentic strategy to process user-provided tasks and database information. +- Sets up database query configurations through the `DatabaseQueryTool`. +- Employs the `mustache_render` utility to tailor command-line interactions and strategy prompts. +- Executes the task, restricting output operation to a maximum of ten results (`limit=10`). + +### `run` Method + +The `run` function executes the agent's strategy to fetch requested data and append execution usage metrics from the `AgenticStrategyV2`. + +By employing this module, users can expect an automated task execution environment, where high-level natural language prompts can seamlessly translate into concrete database operations. This setup is particularly useful for data analysis and management tasks that require flexible and interactive database querying capabilities. diff --git a/patchwork/steps/GitHubAgent/GitHubAgent.py b/patchwork/steps/GitHubAgent/GitHubAgent.py index bc8d319c1..0ac014538 100644 --- a/patchwork/steps/GitHubAgent/GitHubAgent.py +++ b/patchwork/steps/GitHubAgent/GitHubAgent.py @@ -5,6 +5,7 @@ AgentConfig, AgenticStrategyV2, ) +from patchwork.common.tools.git_tool import GitTool from patchwork.common.tools.github_tool import GitHubTool from patchwork.common.utils.utils import mustache_render from patchwork.step import Step @@ -34,7 +35,10 @@ def __init__(self, inputs): AgentConfig( name="Assistant", model="gemini-2.0-flash", - tool_set=dict(github_tool=GitHubTool(base_path, inputs["github_api_key"])), + tool_set=dict( + github_tool=GitHubTool(base_path, inputs["github_api_key"]), + git_tool=GitTool(base_path), + ), system_prompt="""\ You are a senior software developer helping the program manager to obtain some data from GitHub. You can access github through the `gh` CLI app. diff --git a/pyproject.toml b/pyproject.toml index ab89dee54..d136584c5 100644 --- a/pyproject.toml +++ b/pyproject.toml @@ -1,6 +1,6 @@ [tool.poetry] name = "patchwork-cli" -version = "0.0.123" +version = "0.0.124" description = "" authors = ["patched.codes"] license = "AGPL"