Skip to content

docs: code agent fixes #593

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Merged
merged 3 commits into from
Feb 20, 2025
Merged

docs: code agent fixes #593

Changes from all commits
Commits
Show all changes
3 commits
Select commit Hold shift + click to select a range
90fdbb3
.
jayhack Feb 20, 2025
97bd29e
.
jayhack Feb 20, 2025
d9d3a55
.
jayhack Feb 20, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
243 changes: 90 additions & 153 deletions docs/tutorials/build-code-agent.mdx
View file
Open in desktop
Original file line number Diff line number Diff line change
@@ -1,205 +1,146 @@
---
title: "Building a Code Agent with LangChain"
title: "Building Code Agents"
sidebarTitle: "Code Agent"
icon: "robot"
iconType: "solid"
---

This guide demonstrates how to build an intelligent code agent that can analyze and manipulate codebases using Codegen's LangChain integration.
This guide demonstrates how to build an intelligent code agent that can analyze and manipulate codebases.

This agent access to powerful code viewing and manipulation tools powered by Codegen, including:
- `RevealSymbolTool`: reveal all N-th degree dependencies and usages of a function
- `MoveSymbolTool`: move a symbol between files, updating all imports etc. (guaranteed correctness)
- `SemanticEditTool`: implementation of Cursor-style smart file editing
- `SemanticSearchTool`: search over an index of vector embeddings for files
```python
from codegen import CodeAgent, Codebase

# Grab a repo from Github
codebase = Codebase.from_repo('fastapi/fastapi')

<Info>View the full code for the default tools and agent implementation in our [examples repository](https://github.com/codegen-sh/codegen-sdk/tree/develop/src/codegen/extensions/langchain)</Info>
# Create a code agent with read/write codebase access
agent = CodeAgent(codebase)

# Run the agent with a prompt
agent.run("Tell me about this repo")
```

# Agent Starter Code
The agent has access to powerful code viewing and manipulation tools powered by Codegen, including:
- `ViewFileTool`: View contents and metadata of files
- `SemanticSearchTool`: Search over code using natural language queries
- `SemanticEditTool`: Make intelligent edits to files
- `RevealSymbolTool`: Analyze symbol dependencies and usages
- `MoveSymbolTool`: Move symbols between files with import handling
- And many more...

Below is a code snippet you can run as-is to solve SWE bench issues.
<Info>View the full code for the default tools and agent implementation in our [examples repository](https://github.com/codegen-sh/codegen-sdk/tree/develop/src/codegen/extensions/langchain/tools)</Info>

# Basic Usage

The following example shows how to create and run a `CodeAgent`:

```python
from codegen import Codebase
from codegen.extensions.langchain.agent import create_agent_with_tools
from codegen.extensions.langchain.tools import (
CreateFileTool,
DeleteFileTool,
EditFileTool,
ListDirectoryTool,
MoveSymbolTool,
RenameFileTool,
RevealSymbolTool,
SearchTool,
SemanticEditTool,
ViewFileTool,
)
from codegen import CodeAgent, Codebase

# Parse Codebase
# Using a repo + commit from SWE-Bench
codebase = Codebase.from_repo('sqlfluff/sqlfluff', commit='a820c139ccbe6d1865d73c4a459945cd69899f8f', language='python')

# Define tools
tools = [
ViewFileTool(codebase), # View file contents
ListDirectoryTool(codebase), # List directory contents
SearchTool(codebase), # Search code
EditFileTool(codebase), # Edit files
CreateFileTool(codebase), # Create new files
DeleteFileTool(codebase), # Delete files
RenameFileTool(codebase), # Rename files
MoveSymbolTool(codebase), # Move functions/classes
RevealSymbolTool(codebase), # Analyze symbol relationships
SemanticEditTool(codebase), # Make semantic edits
]


agent = create_agent_with_tools(codebase, tools)

# Using a prompt from SWE Bench
issue = """Enable quiet mode/no-verbose in CLI for use in pre-commit hook There seems to be only an option to increase the level of verbosity when using SQLFluff [CLI](https://docs.sqlfluff.com/en/stable/cli.html), not to limit it further. It would be great to have an option to further limit the amount of prints when running `sqlfluff fix`, especially in combination with deployment using a pre-commit hook. For example, only print the return status and the number of fixes applied, similar to how it is when using `black` in a pre-commit hook: ![image](https://user-images.githubusercontent.com/10177212/140480676-dc98d00b-4383-44f2-bb90-3301a6eedec2.png) This hides the potentially long list of fixes that are being applied to the SQL files, which can get quite verbose."""
prompt = f"""
Hey CodegenBot!

Here's a SWE task for you. Please solve this task diligently.

>>>>> ISSUE TEXT <<<<<
{issue}
>>>>> ISSUE TEXT <<<<<

Godspeed, CodegenBot! 🚀
"""

# Run the agent
result = agent.invoke(
{'input': prompt},
config={"configurable": {"session_id": "demo"}}
)
# Grab a repo from Github
codebase = Codebase.from_repo('fastapi/fastapi')

# Create a code agent with read/write codebase access
agent = CodeAgent(codebase)

# Run the agent with a prompt
agent.run("Tell me about this repo")
```
<Note>Your `ANTHROPIC_API_KEY` must be set in your env.</Note>

## Step 1: Tool Configuration
# Available Tools

The agent comes with several built-in tools for code operations:
The agent comes with a comprehensive set of tools for code analysis and manipulation. Here are some key tools:

```python
from codegen.extensions.langchain.tools import (
ViewFileTool,
SemanticSearchTool,
SemanticEditTool,
RevealSymbolTool,
MoveSymbolTool,
EditFileTool,
CreateFileTool,
DeleteFileTool,
EditFileTool,
ListDirectoryTool,
MoveSymbolTool,
RenameFileTool,
RevealSymbolTool,
SearchTool,
SemanticEditTool,
ViewFileTool,
)

tools = [
ViewFileTool(codebase), # View file contents
ListDirectoryTool(codebase), # List directory contents
SearchTool(codebase), # Search code
EditFileTool(codebase), # Edit files
CreateFileTool(codebase), # Create new files
DeleteFileTool(codebase), # Delete files
RenameFileTool(codebase), # Rename files
MoveSymbolTool(codebase), # Move functions/classes
RevealSymbolTool(codebase), # Analyze symbol relationships
SemanticEditTool(codebase), # Make semantic edits
]
```

## Step 2: Setting Up the Agent
<Note>View the full set of [tools on Github](https://github.com/codegen-sh/codegen-sdk/blob/develop/src/codegen/extensions/langchain/tools.py)</Note>

You can create an agent using `create_agent_with_tools`:
Each tool provides specific capabilities:

```python
from langchain_openai import ChatOpenAI
from codegen import Codebase
from codegen.extensions.langchain.agent import create_agent_with_tools

# Create the agent with GPT-4
agent = create_agent_with_tools(codebase=codebase, tools=tools)
```

The agent is initialized with:
- A Codebase instance to operate on
- An LLM (GPT-4o in this case)
- Tools for code manipulation
- A conversation memory to maintain context
# Extensions

## GitHub Integration

The agent includes tools for GitHub operations like PR management. Set up GitHub access with:

Each tool provides specific capabilities to the agent, allowing it to perform complex code operations.

## Step 3: Interacting with the Agent
```bash
CODEGEN_SECRETS__GITHUB_TOKEN="..."
```

Let's see some examples of how to interact with the agent:
Import the GitHub tools:

```python
# Analyze dependencies
result = agent.invoke(
{"input": "What are the dependencies of the FastAPI class?"},
config={"configurable": {"session_id": "demo"}}
from codegen.extensions.langchain.tools import (
GithubCreatePRTool,
GithubViewPRTool,
GithubCreatePRCommentTool,
GithubCreatePRReviewCommentTool
)
print(result["output"])
```

# Find usage patterns
result = agent.invoke(
{"input": "Show me examples of dependency injection in the codebase"},
config={"configurable": {"session_id": "demo"}}
)
print(result["output"])
These tools enable:
- Creating pull requests
- Viewing PR contents and diffs
- Adding general PR comments
- Adding inline review comments

# Perform code analysis
result = agent.invoke(
{"input": "What's the most complex function in terms of dependencies?"},
config={"configurable": {"session_id": "demo"}}
)
print(result["output"])
```
<Note>View all Github tools on [Github](https://github.com/codegen-sh/codegen-sdk/blob/develop/src/codegen/extensions/langchain/tools.py)</Note>

The agent maintains conversation history, so it can reference previous queries and build context over time.

## Step 4: Code Manipulation
## Linear Integration

The agent can also perform code changes:
The agent can interact with Linear for issue tracking and project management. To use Linear tools, set the following environment variables:

```python
# Move a function to a new file
result = agent.invoke(
{"input": "Move the validate_email function to validation_utils.py"},
config={"configurable": {"session_id": "demo"}}
)
```bash
LINEAR_ACCESS_TOKEN="..."
LINEAR_TEAM_ID="..."
LINEAR_SIGNING_SECRET="..."
```

# Rename a class and update all references
result = agent.invoke(
{"input": "Rename the UserModel class to User and update all imports"},
config={"configurable": {"session_id": "demo"}}
)
Import and use the Linear tools:

# Add error handling
result = agent.invoke(
{"input": "Add proper error handling to the process_data function"},
config={"configurable": {"session_id": "demo"}}
```python
from codegen.extensions.langchain.tools import (
LinearGetIssueTool,
LinearGetIssueCommentsTool,
LinearCommentOnIssueTool,
LinearSearchIssuesTool,
LinearCreateIssueTool,
LinearGetTeamsTool
)
```

The agent will:
1. Analyze the current code state
2. Plan the necessary changes
3. Execute the changes while maintaining code correctness
4. Update all related imports and references
These tools allow the agent to:
- Create and search issues
- Get issue details and comments
- Add comments to issues
- View team information

<Note>View all Linear tools on [Github](https://github.com/codegen-sh/codegen-sdk/blob/develop/src/codegen/extensions/langchain/tools.py)</Note>

## Advanced Usage

### Adding Custom Tools
## Adding Custom Tools

You can extend the agent with custom tools:

```python
from langchain.tools import BaseTool
from pydantic import BaseModel, Field
from codegen import CodeAgent

class CustomToolInput(BaseModel):
"""Input schema for custom tool."""
Expand All @@ -217,9 +158,5 @@ class CustomCodeTool(BaseTool):

# Add custom tool to agent
tools.append(CustomCodeTool())
agent = create_codebase_agent(
codebase=codebase,
tools=tools,
model_name="gpt-4o"
)
agent = CodebaseAgent(codebase, tools=tools, model_name="gpt-4o")
```