Currently, when working in a monorepo or a multi-project architecture (e.g., a C# Solution), codegraph and claude code restrict their context and file resolution to the current working directory (CWD). I would like to request a feature that allows developers to define external scan paths or specify a "workspace root" while keeping the CLI execution focused on a specific subdirectory.
My Use Case / Project Structure
I have a C# project with the following directory structure:
Program/
├── Shared/ # Shared DTOs, utilities, and interfaces
├── Client/ # Client-side application
└── Server/
└── LogicServer/ # The core service I am currently developing (CWD)
My LogicServer.sln references the Shared project. I want to run claude inside Server/LogicServer/ so that the AI focuses solely on writing and modifying my server logic, but it still needs to understand the classes and references from ../../Shared/.
The Current Problem
Running in CWD: If I run claude inside LogicServer, it has no knowledge of Shared/ and hallucinates external dependencies.
Path Misalignment: If I run codegraph init at the Program/ root, but start claude inside LogicServer/, the relative paths in the graph break, leading to file-not-found errors.
Loss of Sandbox Focus: The workaround is to start claude at the Program/ root and try to restrict it using prompts. However, this often fails—the AI gets distracted by the broader project scope and sometimes attempts to modify Client/ or root files, breaking the sandbox experience.
Proposed Solution
It would be extremely helpful to decouple the Working Directory (Sandbox) from the Context/Scan Boundary.
Here are two potential ways to implement this:
Allow passing include paths when initializing or running.
codegraph init --include ../../Shared
Currently, when working in a monorepo or a multi-project architecture (e.g., a C# Solution), codegraph and claude code restrict their context and file resolution to the current working directory (CWD). I would like to request a feature that allows developers to define external scan paths or specify a "workspace root" while keeping the CLI execution focused on a specific subdirectory.
My Use Case / Project Structure
I have a C# project with the following directory structure:
My LogicServer.sln references the Shared project. I want to run claude inside Server/LogicServer/ so that the AI focuses solely on writing and modifying my server logic, but it still needs to understand the classes and references from ../../Shared/.
The Current Problem
Running in CWD: If I run claude inside LogicServer, it has no knowledge of Shared/ and hallucinates external dependencies.
Path Misalignment: If I run codegraph init at the Program/ root, but start claude inside LogicServer/, the relative paths in the graph break, leading to file-not-found errors.
Loss of Sandbox Focus: The workaround is to start claude at the Program/ root and try to restrict it using prompts. However, this often fails—the AI gets distracted by the broader project scope and sometimes attempts to modify Client/ or root files, breaking the sandbox experience.
Proposed Solution
It would be extremely helpful to decouple the Working Directory (Sandbox) from the Context/Scan Boundary.
Here are two potential ways to implement this:
Allow passing include paths when initializing or running.
codegraph init --include ../../Shared