Skip to main content

Documentation Index

Fetch the complete documentation index at: https://mintlify.com/swe-agent/mini-swe-agent/llms.txt

Use this file to discover all available pages before exploring further.

mini is a REPL-style interactive command-line interface for running mini-swe-agent directly in your local environment. It is designed for interactive, exploratory use — you can guide the agent, confirm or reject its actions, take the wheel yourself, and steer it toward a solution without leaving the terminal. When you run mini, you are prompted for a task (or pass one with -t). The agent then proposes actions in a loop, and depending on the mode you choose, those actions are either executed immediately, held for your approval, or typed by you directly.

CLI flags

FlagDescription
-t, --taskThe task or problem statement to give the agent. If omitted, you will be prompted interactively.
-c, --configPath, filename, or key-value pair for the config to use. Defaults to mini.yaml or the path set in MSWEA_MINI_CONFIG_PATH. Multiple -c flags are merged recursively.
-m, --modelModel name to use (e.g., anthropic/claude-sonnet-4-5-20250929). Overrides MSWEA_MODEL_NAME.
-y, --yoloStart in yolo mode — the agent executes actions without asking for confirmation.
-o, --outputPath for the output trajectory file. Defaults to last_mini_run.traj.json in the global config directory.
-l, --cost-limitPer-run cost limit in dollars. Set to 0 to disable.
--environment-classEnvironment backend to use (e.g., local, docker). See Execution environments.
--model-classModel class to use (e.g., litellm, openrouter, or a full import path). Advanced.
--agent-classAgent class to use (e.g., interactive, or a full import path). Advanced.
--exit-immediatelyExit without prompting when the agent wants to finish. Advanced.
-h, --helpShow help and exit.
When you pass -c, the default mini.yaml config is not automatically included. If you want to layer overrides on top of the default config, pass it explicitly: -c mini.yaml -c model.model_kwargs.temperature=0.5.

Modes of operation

mini supports three modes that control who executes commands. You can switch between them at any prompt with a slash command.
In confirm mode, the agent proposes an action and waits for you to approve or reject it before anything runs.When the agent proposes an action, you will see a prompt like:
Execute 1 action(s)? Enter to confirm, type comment to reject, or /h to show available commands
>
  • Press Enter to execute the action.
  • Type a message and press Enter to reject the action and send your comment back to the agent.
  • Type /y to confirm and switch to yolo mode in one step.
  • Type /u to skip execution and switch to human mode.
This is the default mode when you run mini. It gives you full oversight of every command before it runs.

Switching modes mid-session

You can switch modes at any point when the agent is waiting for input. The following slash commands are available at every prompt:
CommandAction
/ySwitch to yolo mode
/cSwitch to confirm mode
/uSwitch to human mode
/mEnter a multiline comment
/hShow available commands and current mode

Interrupting the agent

Press Ctrl+C at any time to interrupt the agent mid-step. After interrupting, you will be prompted to type a comment or a slash command. You can:
  • Type a message to send feedback to the agent and resume.
  • Type /y, /c, or /u to switch modes and continue.
  • Press Enter with no input to resume without sending a message.

Trajectory output

mini saves the full history of each run — all messages, costs, and the final exit status — to a trajectory file. By default this is saved to last_mini_run.traj.json inside your global config directory. The path to this directory is printed when you start mini. To save the trajectory to a custom location, use the -o/--output flag:
mini --task "Fix the bug in app.py" --output ./my-run.traj.json
See Trajectories and output files for details on the file format and how to browse trajectories with the inspector.

Build docs developers (and LLMs) love