Work from the terminal
Use the Python CLI for an interactive session or a one-shot task in a selected project directory.
The current entry point
The Python CLI starts the local agent loop directly. It is separate from the server-backed web and desktop interfaces. Run it from the repository root using the project's configured Python environment.
python3 -m agent_system.cli --workspace /path/to/projectInteractive or one-shot
Omit --prompt to keep an interactive session open. Supply --prompt when you want to send one task from the command line. Keep the request focused enough that you can inspect its outcome.
python3 -m agent_system.cli --workspace /path/to/project \
--prompt "trace the failing test and explain the possible cause"Read the working record
Tool activity tells you what was attempted and what came back. A possible cause is an investigation result; a verified fix needs checks that exercise the requested behavior. Treat a failed or unavailable check as unfinished evidence, not as a passing result.
A shell command and a file tool can have different reach. The selected workspace does not make every possible command safe. Review the active policy before approving actions.
Names and compatibility
The public project is Aditya. Some internal modules and package identifiers still use Aroxa. Use the entry point and flags documented for the interface you are running; do not assume a differently named alias or --cwd flag applies to the Python CLI.