repository·main·Indexed 42 days ago
https://github.com/augmentcode/auggieAn agentic coding CLI for the terminal that assists developers with codebase analysis, safe edits, and routine task automation using natural language processing. It features Augment Commands for common development tasks, a Python SDK for Direct Context API-based indexing, and a GitHub Action Repository Indexer for automated semantic search and AI-powered Q&A.
The Context SDK provides several modes for semantic search and codebase analysis:
The FileSystem Context implementation provides several key capabilities for interacting with local workspaces:
search(): Performs semantic search over local files.search_and_ask(): A single-step method for AI-powered Q&A about your workspace.The user_examples directory provides executable scripts demonstrating the core capabilities of the Augment Python SDK. Use these examples to learn about:
01_quick_start.py).02_event_listener_builtin.py) or creating custom event listeners (03_event_listener_custom.py).04_session_management.py).05_type_inference.py) and explicit type specification using dataclasses (06_typed_returns.py).07_success_criteria.py).08_function_tools.py).The FileSystem Context implementation provides several core capabilities for interacting with local codebases:
search(): Performs semantic search over local files.searchAndAsk(): A single-step operation for AI-powered Q&A regarding the workspace.The indexer provides efficient, incremental updates to your repository's semantic index. It follows this lifecycle:
.augment-index-state/{branch}/state.json).When implementing a custom listener, keep the following constraints in mind:
on_tool_call event will have one or more corresponding on_tool_response events sharing the same tool_call_id.on_agent_message_chunk is for real-time streaming (e.g., typewriter effects), while on_agent_message is for the final, complete payload.The Augment Python SDK is divided into two primary functional areas:
FileSystemContext and DirectContext to interact with codebases.The Direct Context implementation provides several key capabilities for interacting with indexed codebases:
search(): Performs semantic searches that return formatted code snippets.searchAndAsk(): A high-level method for one-step AI-powered Q&A based on the indexed code.By default, agent calls are independent. To maintain conversation continuity where subsequent calls remember previous context, use the agent.session() context manager.
with agent.session() as session:
session("Step 1")
session("Step 2 that builds on step 1")
session("Step 3 that uses both")with agent.session() as session:
session("Step 1")
session("Step 2 that builds on step 1")The ClaudeCodeACPClient maintains context automatically across multiple messages sent to the same client instance. This allows for multi-turn conversations where the agent remembers previous instructions and information.
To start a completely new session and wipe the existing conversation history, use the clear_context() method.
client = ClaudeCodeACPClient()
client.start()
# Context is preserved between these calls
client.send_message("My favorite color is blue")
response = client.send_message("What is my favorite color?")
# Clear context to start fresh
client.clear_context()
client.stop()For the agent to correctly understand and invoke your functions, they must adhere to these three requirements:
param: str).Args: section.-> dict).Supported Types include:
str, int, float, boollist, dict, List[T], Dict[K, V]Optional[T]Union[T1, T2]AnyThe converter translates natural language patterns into specific SDK patterns. When writing or reviewing generated code, understand these five core patterns:
agent.session() to maintain continuity between steps.bool) to drive if/else control flow.dataclasses and type hints to pass structured data between steps.agent.run() to allow the agent to execute local Python functions.# Example of Sequential Stages with Context
with agent.session() as session:
session("Step 1: Create the base structure")
session("Step 2: Add features")
# Example of Conditional Logic
result = agent("Check if the file exists", bool)
if result:
agent("Process the existing file")
# Example of Data Dependencies
@dataclass
class FileInfo:
path: str
size: int
files = agent("Analyze all config files", list[FileInfo])
for file in files:
if file.size > 1000:
agent(f"Optimize {file.path}")
# Example of Function Calling
def run_tests(test_file: str) -> dict:
return {"passed": 10, "failed": 0}
result = agent.run(
"Run all tests",
return_type=dict,
functions=[run_tests]
)