mini-swe-agent

https://github.com/swe-agent/mini-swe-agent
Admin
mini-swe-agent is a 100-line Python AI agent designed to solve GitHub issues and other coding tasks,...
Tokens:35,799
Snippets:348
Trust Score:8.8
License:MIT
Update:1 month ago
Show doc for...
Tokens:
Raw
### Quickstart: mini-extra programbench --help and example run

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/usage/programbench.md

Shows how to view help for the programbench runner and run it with a model and 4 workers. Requires installing programbench first (pip install programbench or uvx programbench).

```bash
mini-extra programbench --help
# or
python src/minisweagent/run/benchmarks/programbench.py --help
# Example:
mini-extra programbench \
    --model anthropic/claude-sonnet-4-5-20250929 \
    --workers 4
```

--------------------------------

### Quickstart Guide

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/reference/models/litellm.md

For guidance on setting up local models, refer to the quickstart guide.

```markdown
For local models, please check the [quickstart guide](../../quickstart.md).
```

--------------------------------

### Install from source (developer setup)

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/README.md

Clones the repository and installs in editable mode for development. Requires git and pip; run 'mini' after installation.

```bash
git clone https://github.com/SWE-agent/mini-swe-agent.git
cd mini-swe-agent && pip install -e .
mini  # run the CLI
```

--------------------------------

### Install CLI and Python bindings in current environment

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/README.md

Installs the package and CLI into the current environment. After installation, run the 'mini' command to start the CLI.

```bash
pip install mini-swe-agent
mini  # run the CLI
```

--------------------------------

### Single instance mode help and examples

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/usage/swebench.md

Shows how to get help for the single-instance script and run it on one task instance for debugging. The first example uses an instance ID, the second uses an instance index. Add --exit-immediately to skip the confirmation prompt at exit.

```bash
mini-extra swebench-single --help
# or
python src/minisweagent/run/benchmarks/swebench_single.py --help
# Example:
mini-extra swebench-single \
    --subset verified \
    --split test \
    --model anthropic/claude-sonnet-4-5-20250929 \
    -i sympy__sympy-15599
# or
mini-extra swebench-single \
    --subset verified \
    --split test \
    -m anthropic/claude-sonnet-4-5-20250929 \
    -i 0  # instance index
```

--------------------------------

### Install mini-swe-agent from source

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/quickstart.md

Clone the repository and install in editable mode for development or customization.

```bash
git clone https://github.com/SWE-agent/mini-swe-agent.git
cd mini-swe-agent
pip install -e .
```

--------------------------------

### Install full dependencies

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/reference/environments/swerex_modal.md

Install the necessary dependencies for the mini-swe-agent including full support.

```bash
pip install "mini-swe-agent[full]"
```

--------------------------------

### Batch mode help and example

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/usage/swebench.md

Shows how to get help for the batch mode script and run it on multiple task instances in parallel. The example uses the anthropic/claude-sonnet-4-5-20250929 model, the verified subset, the test split, and 4 workers.

```bash
mini-extra swebench --help
# or
python src/minisweagent/run/benchmarks/swebench.py --help
# Example:
mini-extra swebench \
    --model anthropic/claude-sonnet-4-5-20250929 \
    --subset verified \
    --split test \
    --workers 4
```

--------------------------------

### Permanently install mini-swe-agent with uv

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/quickstart.md

Install the CLI permanently using uv tool install, then run the commands.

```bash
uv tool install mini-swe-agent
# then
mini  # CLI
mini-extra  # extra utilities
```

--------------------------------

### Run global configuration setup

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/global_configuration.md

Use this command to interactively set the default model and API keys.

```bash
mini-extra config setup
```

--------------------------------

### Set up Modal authentication

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/reference/environments/swerex_modal.md

Run the Modal setup command to authenticate your account.

```bash
modal setup
```

--------------------------------

### Run mini-swe-agent from source

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/quickstart.md

After installing from source, run the CLI and extra utilities.

```bash
mini  # CLI
mini-extra  # extra utilities
```

--------------------------------

### Install mini-swe-agent with pip

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/quickstart.md

Use pip to install the package in your current environment. After installation, you can run the CLI with 'mini' and extra utilities with 'mini-extra'.

```bash
pip install mini-swe-agent
```

```bash
mini  # CLI
mini-extra  # extra utilities
```

--------------------------------

### Permanently install mini-swe-agent with pipx

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/quickstart.md

Install the CLI permanently using pipx install, then run the commands. If the invocation doesn't immediately work, you might need to run 'pipx ensurepath'.

```bash
pipx install mini-swe-agent
# then
mini  # CLI
mini-extra  # extra utilities
```

--------------------------------

### Install dev dependencies and pre-commit hooks

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/quickstart.md

For contributors, install dev dependencies and pre-commit hooks.

```bash
pip install -e '.[dev]'
pip install pre-commit && pre-commit install
```

--------------------------------

### Install full package with extra dependencies

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/quickstart.md

Install the maximal package including extra dependencies.

```bash
pip install -e '.[full]'
```

--------------------------------

### Install CLI via uvx or pipx

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/README.md

Quickly try the mini-swe-agent CLI in an anonymous virtual environment. Use either uvx or pipx; pipx requires ensurepath before running.

```bash
pip install uv && uvx mini-swe-agent
# or
pip install pipx && pipx ensurepath && pipx run mini-swe-agent
```

--------------------------------

### Install ConTree dependencies

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/reference/environments/contree.md

Install the necessary dependencies to use the ConTree environment with mini-swe-agent.

```bash
pip install "mini-swe-agent[contree]"
```

--------------------------------

### Hello world example with DefaultAgent

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/usage/python_bindings.md

Basic usage of mini-SWE-agent as a Python library. Sets up logging, creates a DefaultAgent with a specified model and a LocalEnvironment, then runs the agent on a task. The model name is given as 'anthropic/claude-sonnet-4-5-20250929'.

```python
import logging

from minisweagent.agents.default import DefaultAgent
from minisweagent.models import get_model
from minisweagent.environments.local import LocalEnvironment

logging.basicConfig(level=logging.DEBUG)
task = "Write a hello world program"
model_name = "anthropic/claude-sonnet-4-5-20250929"

agent = DefaultAgent(
    get_model(input_model_name=model_name),
    LocalEnvironment(),
)

# Run the agent
agent.run(task)
```

--------------------------------

### Default agent example

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/cookbook.md

Shows how to create a DefaultAgent using get_model for automatic model selection and a LocalEnvironment. The model_name variable is assumed to be defined elsewhere.

```python
from minisweagent.agents.default import DefaultAgent
from minisweagent.models import get_model
from minisweagent.environments.local import LocalEnvironment

agent = DefaultAgent(
    get_model(input_model_name=model_name),
    LocalEnvironment(),
)
```

--------------------------------

### Launch vLLM server

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/models/local_models.md

Starts the vLLM server with the specified model, making it available at http://localhost:8000 by default.

```bash
vllm serve ricdomolm/mini-coder-1.7b &
```

--------------------------------

### Run mini-swe-agent with uvx (isolated)

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/quickstart.md

Use uvx to run the CLI without installing Python bindings. The package is installed in an anonymous virtual environment.

```bash
uvx mini-swe-agent  # CLI
uvx --from mini-swe-agent mini-extra  # extra utilities
```

--------------------------------

### Local evaluation with SWE-bench harness

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/usage/swebench.md

Runs evaluation locally using the SWE-bench package. Requires a local installation of SWE-bench. Replace <num_workers> and <run_id> with your values.

```bash
python -m swebench.harness.run_evaluation \
    --dataset_name princeton-nlp/SWE-bench_Verified \
    --predictions_path preds.jsonl \
    --max_workers <num_workers> \
    --run_id <run_id>
```

--------------------------------

### Cloud-based evaluation with sb-cli

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/usage/swebench.md

Submits predictions to SWE-bench's free cloud-based evaluation service. Requires installing sb-cli and obtaining a token. Results typically arrive within 20 minutes.

```bash
sb-cli submit swe-bench_verified test --predictions_path preds.json --run_id some-id-for-your-run
```

--------------------------------

### Run mini-swe-agent with pipx (isolated)

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/quickstart.md

Use pipx to run the CLI without installing Python bindings. The package is installed in an anonymous virtual environment.

```bash
# CLI
pipx run mini-swe-agent
# Extra utilities
pipx run --spec mini-swe-agent mini-extra
```

--------------------------------

### Use instance variables in env_startup_command with jinja2

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/usage/swebench.md

The env_startup_command is rendered with instance variables as template variables using jinja2. This example clones a repository using the repo_url variable, which is useful with environments like bubblewrap.

```yaml
run:
  env_startup_command: "git clone {{ repo_url }} . --force"
```

--------------------------------

### Message examples in trajectory file (JSON)

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/usage/output_files.md

Shows examples of the different message types in a trajectory file: system, user, assistant, observation, and final submission. Assistant messages include parsed actions in the extra field. Observation messages use role 'user' with returncode and output. Final messages contain exit_status and submission.

```json
// System message (agent instructions)
{"role": "system", "content": "You are a helpful assistant..."}

// User message (task description)
{"role": "user", "content": "Please solve this issue: ..."}

// Assistant message (model response with parsed actions)
{
  "role": "assistant",
  "content": "Let me check the files...\n\n```mswea_bash_command\nls -la\n```",
  "extra": {
    "actions": [{"command": "ls -la"}],  // parsed actions to execute
    "cost": 0.003,  // cost of this API call
    "timestamp": 1706000000.0,  // unix timestamp of when this message was created
    "response": { ... }  // raw API response
  }
}

// Observation message (execution result)
{
  "role": "user",
  "content": "<returncode>0</returncode>\n<output>\nfile1.py\nfile2.py\n</output>",
  "extra": {
    "returncode": 0,
    "timestamp": 1706000001.0
  }
}

// Final message (when agent submits)
{
  "role": "user",
  "content": "",
  "extra": {
    "exit_status": "Submitted",
    "submission": "diff --git a/file.py..."
  }
}
```

--------------------------------

### Set a startup command in the environment with run.env_startup_command

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/usage/swebench.md

Use the run.env_startup_command config option to run a command in the environment before the agent starts. The command is rendered with instance variables as template variables using jinja2.

```yaml
run:
  env_startup_command: "apt-get update && apt-get install -y python3-pip"
```

--------------------------------

### Human in the loop agent example

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/cookbook.md

Shows how to create an InteractiveAgent for human-in-the-loop interactions, using get_model for automatic model selection and a LocalEnvironment. The model_name variable is assumed to be defined elsewhere.

```python
from minisweagent.agents.interactive import InteractiveAgent
from minisweagent.models import get_model
from minisweagent.environments.local import LocalEnvironment

agent = InteractiveAgent(
    get_model(input_model_name=model_name),
    LocalEnvironment(),
)
```

--------------------------------

### Popular model name examples

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/models/quickstart.md

A list of commonly used model names formatted with their respective providers.

```text
anthropic/claude-opus-4-6-20260205
openai/gpt-5.4
openai/gpt-5.4-mini
gemini/gemini-2.5-pro
deepseek/deepseek-chat
```

--------------------------------

### XML format configuration for action parsing

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/yaml_configuration.md

This example shows a YAML configuration that uses XML tags (<action>) instead of markdown code blocks for action parsing. It is based on the default mini.yaml config and can be loaded with --config swebench_xml.

```yaml
--8<-- "src/minisweagent/config/benchmarks/swebench_xml.yaml"
```

--------------------------------

### Usage: Start inspector and open trajectory files

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/usage/inspector.md

Commands to launch the inspector, either by searching for .traj.json files recursively from the current directory or by specifying a specific file or directory.

```bash
# Find all .traj.json files recursively from current directory
mini-extra inspector
# or shorter
mini-e i
# Open the inspector for a specific file
mini-e i <path_to_traj.json>
# Search for trajectory files in a specific directory
mini-e i <path_to_directory>
```

--------------------------------

### Environment._check_finished method

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/control_flow.md

Checks if the command output starts with the magic string 'COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT' and raises a Submitted exception with an exit message.

```python
# In Environment.execute
def _check_finished(self, output: dict):
    lines = output.get("output", "").lstrip().splitlines(keepends=True)
    if lines and lines[0].strip() == "COMPLETE_TASK_AND_SUBMIT_FINAL_OUTPUT":
        raise Submitted({"role": "exit", "content": ..., "extra": {...}})
```

--------------------------------

### Remove a configuration key

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/usage/config.md

Remove a key from the global configuration file. The example unset the MSWEA_MODEL_NAME key.

```bash
mini-extra config unset MSWEA_MODEL_NAME
```

--------------------------------

### Set default model via mini-extra config set

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/faq.md

Permanently changes the default model by setting the MSWEA_MODEL_NAME configuration key. The example uses anthropic/claude-sonnet-4-5-20250929.

```bash
mini-extra config set MSWEA_MODEL_NAME anthropic/claude-sonnet-4-5-20250929
```

--------------------------------

### Set configuration key

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/usage/config.md

Set a specific key in the global configuration file, either by providing the key and value in the command or interactively when the value is omitted. The example shows setting MSWEA_MODEL_NAME to 'anthropic/claude-sonnet-4-5-20250929'.

```bash
# example: set default model
mini-extra config set MSWEA_MODEL_NAME anthropic/claude-sonnet-4-5-20250929
# or interactively
mini-extra config set
```

--------------------------------

### Portkey cost calculation error due to missing provider

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/models/troubleshooting.md

Occurs when the Portkey model name does not match the litellm model name and lacks the provider. Fix by setting `litellm_model_name_override` to the litellm model name with provider, as shown in the YAML example.

```text
  File "/opt/miniconda3/envs/clash/lib/python3.10/site-packages/minisweagent/models/portkey_model.py", line 85, in query
    cost = litellm.cost_calculator.completion_cost(response)
  File "/opt/miniconda3/envs/clash/lib/python3.10/site-packages/litellm/cost_calculator.py", line 973, in completion_cost
    raise e
  File "/opt/miniconda3/envs/clash/lib/python3.10/site-packages/litellm/cost_calculator.py", line 966, in completion_cost
    raise e
  File "/opt/miniconda3/envs/clash/lib/python3.10/site-packages/litellm/cost_calculator.py", line 928, in completion_cost
    ) = cost_per_token(
  File "/opt/miniconda3/envs/clash/lib/python3.10/site-packages/litellm/cost_calculator.py", line 218, in cost_per_token
    _, custom_llm_provider, _, _ = litellm.get_llm_provider(model=model)
  File "/opt/miniconda3/envs/clash/lib/python3.10/site-packages/litellm/litellm_core_utils/get_llm_provider_logic.py", line 395, in get_llm_provider
    raise e
  File "/opt/miniconda3/envs/clash/lib/python3.10/site-packages/litellm/litellm_core_utils/get_llm_provider_logic.py", line 372, in get_llm_provider
    raise litellm.exceptions.BadRequestError(  # type: ignore
litellm.exceptions.BadRequestError: litellm.BadRequestError: LLM Provider NOT provided. Pass in the LLM provider you are trying to call. You passed model=grok-code-fast-1
 Pass model as E.g. For 'Huggingface' inference endpoints pass in `completion(model='huggingface/starcoder',..)` Learn more: https://docs.litellm.ai/docs/providers
```

--------------------------------

### Add Python functions by subclassing the agent

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/cookbook.md

Subclass DefaultAgent to intercept actions starting with 'python_function', execute them with shlex-split arguments, and return the formatted observation. Other actions fall through to the default behavior.

```python
from minisweagent.agents.default import DefaultAgent
import shlex

def python_function(*args) -> dict:
    ...
    return {"output": "..."}

class AgentWithPythonFunctions(DefaultAgent):
    def execute_actions(self, message: dict) -> list[dict]:
        for action in message.get("extra", {}).get("actions", []):
            command = action.get("command", "")
            if command.startswith("python_function"):
                args = shlex.split(command.removeprefix("python_function").strip())
                return self.add_messages(self.model.format_observation_messages(
                    message, [python_function(*args)], self.get_template_vars()
                ))
        # everything else works as usual
        return super().execute_actions(message)
```

--------------------------------

### Model not mapped error for cost calculation

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/models/troubleshooting.md

Raised when litellm does not know the cost for the model. Add the model to the model registry as described in the local models guide. Also ensure the provider is included in the model name (e.g., `gemini/gemini-2.0-flash`).

```text
Exception: This model isn't mapped yet. model=together_ai/qwen/qwen3-coder-480b-a35b-instruct-fp8, custom_llm_provider=together_ai.
Add it here - https://github.com/BerriAI/litellm/blob/main/model_prices_and_context_window.json.
```

--------------------------------

### Run a hello world script from source

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/quickstart.md

Run a run script from the source tree.

```bash
python src/minisweagent/run/hello_world.py
```

--------------------------------

### Hello world with automatic model selection

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/cookbook.md

Shows how to create a DefaultAgent using the get_model function for automatic model selection. Requires a model name string and a LocalEnvironment.

```python
from minisweagent.agents.default import DefaultAgent
from minisweagent.models import get_model
from minisweagent.environments.local import LocalEnvironment

model_name = "anthropic/claude-sonnet-4-5-20250929"

agent = DefaultAgent(
    get_model(input_model_name=model_name),
    LocalEnvironment(),
)
agent.run(task)
```

--------------------------------

### Hello world with local execution

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/cookbook.md

Shows how to create a DefaultAgent with a LitellmModel and a LocalEnvironment for local execution. The model_name variable is assumed to be defined elsewhere.

```python
from minisweagent.environments.local import LocalEnvironment

agent = DefaultAgent(
    LitellmModel(model_name=model_name),
    LocalEnvironment(),
)
```

--------------------------------

### Hello world with Litellm model

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/cookbook.md

Shows how to create a DefaultAgent using the LitellmModel class with a specific model name. Requires a model name string and a LocalEnvironment.

```python
from minisweagent.agents.default import DefaultAgent
from minisweagent.models.litellm_model import LitellmModel
from minisweagent.environments.local import LocalEnvironment

model_name = "gpt-4o"

agent = DefaultAgent(
    LitellmModel(model_name=model_name),
    LocalEnvironment(),
)
agent.run(task)
```

--------------------------------

### Hello world with docker execution

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/cookbook.md

Shows how to create a DefaultAgent with a LitellmModel and a DockerEnvironment for containerized execution. The model_name variable is assumed to be defined elsewhere.

```python
from minisweagent.environments.docker import DockerEnvironment

agent = DefaultAgent(
    LitellmModel(model_name=model_name),
    DockerEnvironment(),
)
```

--------------------------------

### CLI usage with multiple configs and key-value overrides

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/v2_migration.md

Shows how to pass multiple config files and override specific keys using -c. Note that using -c skips loading the full default config, so always include your base config file.

```bash
mini -c mini.yaml -c model.model_kwargs.temperature=0.5
mini -c swebench.yaml -c agent.step_limit=100
mini -c mini.yaml -c /path/to/model.yaml
```

--------------------------------

### Configure Portkey model via CLI and config

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/models/quickstart.md

Use the portkey model class via the command line or by setting the model_class key in the agent configuration file.

```bash
mini -m "claude-sonnet-4-5-20250929" --model-class portkey
```

```yaml
model:
    model_name: "claude-sonnet-4-5-20250929"
    model_class: portkey
```

--------------------------------

### Evaluate on SWE-bench using Modal

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/reference/environments/swerex_modal.md

Execute a SWE-bench evaluation using the specified Modal configuration file.

```bash
mini-extra swebench \
    --config src/minisweagent/config/extra/swebench_modal.yaml \
    --subset verified \
    --split test \
    --workers 100 \
    -o ./results/gpt5-mini-modal
```

--------------------------------

### Python bindings: Create and run a DefaultAgent

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/index.md

Shown in the 'Python bindings' section of the overview table. Instantiates a DefaultAgent with a LitellmModel and LocalEnvironment, then runs it with the task 'Write a sudoku game'. The model_name argument is left as a placeholder in the source.

```python
agent = DefaultAgent(
    LitellmModel(model_name=...),
    LocalEnvironment(),
)
agent.run("Write a sudoku game")
```

--------------------------------

### Run commands for tool calling and text-based parsing

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/v2_migration.md

Shows the default CLI commands for tool calling and the commands to enable text-based parsing using the 'swebench_backticks.yaml' and 'mini_textbased.yaml' configs.

```bash
# Default (tool calling)
mini
python -m minisweagent.run.benchmarks.swebench
mini-extra swebench

# Text-based parsing
mini-extra swebench -c swebench_backticks.yaml
mini -c mini_textbased.yaml
```

--------------------------------

### Configure environment executables

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/global_configuration.md

Sets the paths or names for singularity, docker, and bubblewrap executables.

```bash
# Path/name to the singularity/apptainer executable
# (default: "singularity")
MSWEA_SINGULARITY_EXECUTABLE="singularity"

# Path/name to the docker executable
# (default: "docker")
MSWEA_DOCKER_EXECUTABLE="docker"

# Path/name to the bubblewrap executable
# (default: "bwrap")
MSWEA_BUBBLEWRAP_EXECUTABLE="bwrap"
```

--------------------------------

### Configure default config file paths

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/global_configuration.md

Sets custom directories and paths for agent configuration files and inspector styles.

```bash
# Set a custom directory for agent config files in addition to the builtin ones
# This allows to specify them by names
MSWEA_CONFIG_DIR="/path/to/your/own/config/dir"

# Config path for mini run script
# (default: package_dir / "config" / "mini.yaml")
MSWEA_MINI_CONFIG_PATH="/path/to/your/own/config"

# Custom style path for trajectory inspector
# (default: package_dir / "config" / "inspector.tcss")
MSWEA_INSPECTOR_STYLE_PATH="/path/to/your/inspector/style.tcss"
```

--------------------------------

### Set environment variables

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/global_configuration.md

Recommended for temporary experimentation or API keys on Linux/macOS or Windows.

```bash
export KEY="value"
# windows:
setx KEY "value"
```

--------------------------------

### Configure Openrouter model via CLI and config

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/models/quickstart.md

Use the openrouter model class via the command line or by setting the model_class key in the agent configuration file.

```bash
mini -m "moonshotai/kimi-k2-0905" --model-class openrouter
```

```yaml
model:
    model_name: "moonshotai/kimi-k2-0905"
    model_class: openrouter
```

--------------------------------

### Configure default run script

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/global_configuration.md

Sets the default entry point for the main CLI run script.

```bash
# Default run script entry point for the main CLI
# (default: "minisweagent.run.mini")
MSWEA_DEFAULT_RUN="minisweagent.run.mini"
```

--------------------------------

### Run mini-swe-agent with Python bindings

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/README.md

Use this snippet to run a mini-swe-agent instance programmatically. It creates a DefaultAgent with a LitellmModel and a LocalEnvironment, then runs the agent with a task string. The model_name parameter must be set to a valid LiteLLM model identifier.

```python
agent = DefaultAgent(
    LitellmModel(model_name=...),
    LocalEnvironment(),
)
agent.run("Write a sudoku game")

```

--------------------------------

### Run mini-swe-agent with ConTree

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/reference/environments/contree.md

Execute the mini-swe-agent using the ConTree environment class via the command line interface.

```bash
mini-extra swebench \
    --subset verified \
    --split test \
    --workers 100
    --environment-class contree
```

--------------------------------

### Project Directory Structure

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/AGENTS.md

The organization of the mini-SWE-agent codebase.

```bash
minisweagent/__init__  # Protocols/interfaces for all base classes
minisweagent/agents  # Agent control flow & loop
minisweagent/environments  # Executing agent actions
minisweagent/models  # LM interfaces
minisweagent/run  # Run scripts that serve as an entry point
```

--------------------------------

### Create custom registry.json

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/models/local_models.md

Defines model parameters and provider settings in a registry file for the local model.

```bash
cat > registry.json <<'EOF'
{
  "ricdomolm/mini-coder-1.7b": {
    "max_tokens": 40960,
    "input_cost_per_token": 0.0,
    "output_cost_per_token": 0.0,
    "litellm_provider": "hosted_vllm",
    "mode": "chat"
  }
}
EOF
```

--------------------------------

### Configure SWE-bench for local vLLM

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/models/local_models.md

Update the SWE-bench configuration file to point to the local vLLM inference endpoint.

```yaml
model:
  model_name: "hosted_vllm/ricdomolm/mini-coder-1.7b"  # or hosted_vllm/path/to/local/model
  model_kwargs:
    api_base: "http://localhost:8000/v1"  # adjust if using a non-default port/address
```

--------------------------------

### Config implementation file

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/usage/config.md

The full source code for the config utility, referenced in the implementation section. This is a file inclusion marker, so the code content is not available on this page and is excluded.

```python
--8<-- "src/minisweagent/run/utilities/config.py"
```

--------------------------------

### Invoke litellm.completion with model_kwargs

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/models/local_models.md

Shows how mini-swe-agent calls litellm.completion, passing model_name, messages, and model_kwargs. model_kwargs is set in the agent config file to add custom_llm_provider and api_base for local models.

```python
litellm.completion(
    model=model_name,
    messages=messages,
    **model_kwargs
)
```

--------------------------------

### Evaluate output with programbench eval

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/usage/programbench.md

Pass the output directory directly to programbench eval. Each instance writes submission.tar.gz and <instance_id>.traj.json under <output>/<instance_id>/.

```bash
programbench eval <output>/
```

--------------------------------

### Custom config for tool calling and text-based model

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/v2_migration.md

YAML configuration showing the default tool calling model class 'litellm' and the text-based alternative 'litellm_textbased' with an action_regex for extracting bash commands.

```yaml
model:
  model_class: litellm  # tool calling (default)
  model_name: anthropic/claude-sonnet-4-5-20250929

# or for text-based:
model:
  model_class: litellm_textbased
  action_regex: "```mswea_bash_command\s*\n(.*?)\n```"
```

--------------------------------

### Configure LiteLLM Response model via CLI

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/reference/models/litellm_response_toolcall.md

Use the --model-class flag to specify the litellm_response model when running the agent from the command line.

```bash
mini -m "openai/gpt-5.2" --model-class litellm_response
```

--------------------------------

### Exception import changes (old vs new)

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/v2_migration.md

Shows the old import path from minisweagent.agents.default and the new import path from minisweagent.exceptions for Submitted and FormatError.

```python
# Old
from minisweagent.agents.default import Submitted, FormatError

# New
from minisweagent.exceptions import Submitted, FormatError
```

--------------------------------

### Minimax Openrouter configuration

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/models/quickstart.md

Configures Minimax via Openrouter with a temperature of 0.0.

```yaml
model:
    model_name: "minimax/minimax-m2"
    model_class: openrouter
    model_kwargs:
        temperature: 0.0
```

--------------------------------

### GPT 5 mini Portkey configuration

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/models/quickstart.md

Configures GPT 5 mini using the Portkey model class.

```yaml
model:
    model_name: "@openai/gpt-5-mini"
    model_class: portkey
```

--------------------------------

### Deepseek configuration

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/models/quickstart.md

Sets the temperature for the Deepseek reasoner model.

```yaml
model:
    model_name: "deepseek/deepseek-reasoner"
    model_kwargs:
        temperature: 0.0
```

--------------------------------

### Agent.run() return value (old vs new)

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/v2_migration.md

Shows the old tuple return value and the new dict return value for Agent.run(), including how to access submission and exit_status.

```python
# Old (v1)
submission, exit_status = agent.run(task)  # tuple[str, str]

# New (v2)
result = agent.run(task)  # dict
submission = result["submission"]
exit_status = result["exit_status"]
```

--------------------------------

### Environment protocol changes (execute signature and serialize)

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/v2_migration.md

Shows the changed execute() signature (now takes action dict and cwd) and the new serialize() method for custom Environment classes.

```python
# Changed signature
def execute(self, action: dict, cwd: str = "") -> dict[str, Any]: ...  # was: (command: str) -> dict[str, str]

# New method
def serialize(self) -> dict: ...
```

--------------------------------

### get_model

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/reference/models/utils.md

Returns a model instance based on the provided model name and configuration.

```APIDOC
## get_model

### Description
Returns a model instance based on the provided model name and configuration.

### Signature
```python
def get_model(model_name: str, **kwargs) -> Any
```

### Parameters
- **model_name** (str) - Required - The name of the model to instantiate.
- **kwargs** - Optional - Additional configuration parameters passed to the model constructor.

### Returns
- The model instance.

```

--------------------------------

### Default credentials error for missing provider prefix

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/models/troubleshooting.md

Occurs when authenticating with a Gemini model but forgetting to prefix the model name with `gemini/`. Ensure the provider is included in the model name.

```text
  File "/Users/.../.virtualenvs/openai/lib/python3.12/site-packages/google/auth/_default.py", line 685, in default
    raise exceptions.DefaultCredentialsError(_CLOUD_SDK_MISSING_CREDENTIALS)
google.auth.exceptions.DefaultCredentialsError: Your default credentials were not found. To set up Application Default Credentials, see
https://cloud.google.com/docs/authentication/external/set-up-adc for more information.
```

--------------------------------

### Generate SWE-bench trajectory

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/models/local_models.md

Executes the generation process for a specific SWE-bench instance using the custom registry.

```bash
LITELLM_MODEL_REGISTRY_PATH=registry.json mini-extra swebench \
    --output test/ --subset verified --split test --filter '^(django__django-11099)$'
```

--------------------------------

### GPT 5.1 medium Portkey configuration

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/models/quickstart.md

Configures GPT 5.1 medium using the Portkey model class with specified reasoning effort and verbosity.

```yaml
model:
    model_name: "@openai/gpt-5.1"
    model_class: portkey
    model_kwargs:
        reasoning_effort: "medium"
        verbosity: "medium"
```

--------------------------------

### Create EnvironmentWithForbiddenPatterns to block specific shell commands

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/cookbook.md

Extends LocalEnvironment to intercept and block commands matching regex patterns defined in EnvironmentWithForbiddenPatternsConfig. This prevents execution of dangerous commands like rm -rf or sudo password changes.

```python
import re
from minisweagent.agents.default import DefaultAgent
from minisweagent.environments.local import LocalEnvironment, LocalEnvironmentConfig
from minisweagent.models.litellm_model import LitellmModel

class EnvironmentWithForbiddenPatternsConfig(LocalEnvironmentConfig):
    forbidden_patterns: list[str] = [
        r"rm -rf /",
        r"sudo.*passwd",
        r"mkfs.",
    ]

class EnvironmentWithForbiddenPatterns(LocalEnvironment):
    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs, config_class=EnvironmentWithForbiddenPatternsConfig)

    def execute(self, action: dict, cwd: str = "") -> dict:
        command = action.get("command", "")
        for pattern in self.config.forbidden_patterns:
            if re.search(pattern, command, re.IGNORECASE):
                return {"output": "Action blocked: forbidden pattern detected", "returncode": 1}
        return super().execute(action, cwd)

agent = DefaultAgent(
    LitellmModel(model_name=model_name),
    EnvironmentWithForbiddenPatterns(),
)
```

--------------------------------

### Configure global cost and retry limits

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/global_configuration.md

Sets global limits for model calls, costs, and retry attempts.

```bash
# Global limit on number of model calls (0 = no limit)
# (default: 0)
MSWEA_GLOBAL_CALL_LIMIT="100"

# Global cost limit in dollars (0 = no limit)
# (default: 0)
MSWEA_GLOBAL_COST_LIMIT="10.00"

# Number of retry attempts for model API calls
# (default: 10)
MSWEA_MODEL_RETRY_STOP_AFTER_ATTEMPT="10"
```

--------------------------------

### Concise Variable Initialization Style

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/AGENTS.md

Rule 11: Avoid initializing variables just to pass them to a function; pass the expression directly.

```python
# bad
a = func()
Class(a)

# good
Class(func())
```

--------------------------------

### Register extra models

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/global_configuration.md

Specifies the path to a JSON file for registering additional models to litellm.

```bash
LITELLM_MODEL_REGISTRY_PATH="/path/to/your/model/registry.json"
```

--------------------------------

### Local model configuration

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/models/quickstart.md

Configures a local model by specifying the custom_llm_provider and api_base.

```yaml
model:
  model_name: "my-local-model"
  model_kwargs:
    custom_llm_provider: "openai"
    api_base: "https://..."
    ...
```

--------------------------------

### Claude Haiku 4.5 configuration

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/models/quickstart.md

Sets the temperature for Claude Haiku 4.5.

```yaml
model:
    model_name: "anthropic/claude-haiku-4-5-20251001"
    model_kwargs:
        temperature: 0.0
```

--------------------------------

### DefaultAgent.step method

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/control_flow.md

The core method of the agent that queries the model and executes actions. It is called repeatedly by run until the task is finished.

```python
def step(self) -> list[dict]:
    return self.execute_actions(self.query())
```

--------------------------------

### Update specific configuration settings

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/global_configuration.md

Use the set command to update individual configuration keys, such as the model name or API keys.

```bash
mini-extra config set KEY VALUE
# e.g.,
mini-extra config set MSWEA_MODEL_NAME "anthropic/claude-sonnet-4-5-20250929"
mini-extra config set ANTHROPIC_API_KEY "sk-..."
```

--------------------------------

### DefaultAgent.query method

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/control_flow.md

Queries the model for a response based on the current messages, adds the response to the messages, and returns it. Includes a placeholder for limit checks.

```python
def query(self) -> dict:
    # ... limit checks ...
    message = self.model.query(self.messages)
    self.add_messages(message)
    return message
```

--------------------------------

### Set API key permanently via mini-extra config set

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/faq.md

Permanently stores the OPENAI_API_KEY in the config file using the mini-extra CLI. Alternatively, edit the .env file in the config directory.

```bash
mini-extra config set OPENAI_API_KEY sk-test123
```

--------------------------------

### Configure ConTree environment variables

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/reference/environments/contree.md

Set the required environment variables for authentication and connectivity with the ConTree service.

```bash
export CONTREE_TOKEN="your-contree-token"
export CONTREE_BASE_URL="your-given-base-url-for-contree"
```

--------------------------------

### Add Python functions by subclassing the environment

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/cookbook.md

Subclass LocalEnvironment to handle 'python_function' commands directly in the execute method, returning the function result. Other commands use the default execution. The agent is instantiated with the custom environment.

```python
from minisweagent.agents.default import DefaultAgent
from minisweagent.environments.local import LocalEnvironment
import shlex

def python_function(*args) -> dict:
    ...
    return {"output": "..."}

class EnvironmentWithPythonFunctions(LocalEnvironment):
    def execute(self, action: dict, cwd: str = "") -> dict:
        command = action.get("command", "")
        if command.startswith("python_function"):
            args = shlex.split(command.removeprefix("python_function").strip())
            return python_function(*args)
        # all other commands are executed as usual
        return super().execute(action, cwd)

agent = DefaultAgent(
    LitellmModel(model_name=model_name),
    EnvironmentWithPythonFunctions(),
)
```

--------------------------------

### New YAML config structure for v2.0

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/v2_migration.md

Shows the new configuration layout with agent, environment, and model sections. Keys like observation_template and format_error_template moved from agent to model, and timeout_template was removed.

```yaml
agent:
  system_template: "..."
  instance_template: "..."
  step_limit: 0
  cost_limit: 3.
environment:
  cwd: "..."
  timeout: 30
model:
  model_name: "..."
  observation_template: "..."
  format_error_template: "..."
```

--------------------------------

### Edit .env file

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/global_configuration.md

Opens the .env file for direct editing using a helper function.

```bash
mini-extra config edit
```

--------------------------------

### Temperature configuration

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/models/quickstart.md

Sets the temperature for an Anthropic model using the model_kwargs key. Note that temperature support varies by model.

```yaml
model:
  model_name: "anthropic/claude-sonnet-4-5-20250929"
    model_kwargs:
      temperature: 0.0
```

--------------------------------

### Gemini 3 Openrouter configuration

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/models/quickstart.md

Configures Gemini 3 via Openrouter with a temperature of 0.0.

```yaml
model:
    model_name: "google/gemini-3-pro-preview"
    model_class: openrouter
    model_kwargs:
        temperature: 0.0
```

--------------------------------

### Portkey model configuration with litellm_model_name_override

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/models/troubleshooting.md

Configuration to fix Portkey cost calculation errors by manually mapping the litellm model name. Set `litellm_model_name_override` to the litellm model name including the provider (e.g., `xai/grok-code-fast-1`).

```yaml
model:
  model_name: "grok-code-fast-1"  # the portkey model name
  model_class: "portkey"  # make sure to use the portkey model class
  litellm_model_name_override: "xai/grok-code-fast-1"  # the litellm model name for cost information
  ...
```

--------------------------------

### preds.json format (JSON)

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/usage/output_files.md

Shows the structure of the predictions file, which aggregates results from all instances in a SWE-bench compatible format. Each entry is keyed by instance_id and contains model_name_or_path, instance_id, and model_patch.

```json
{
  "owner__repo__123": {  // keyed by instance_id
    "model_name_or_path": "anthropic/claude-sonnet-4-5-20250929",  // model used
    "instance_id": "owner__repo__123",  // SWE-bench instance identifier
    "model_patch": "diff --git a/file.py b/file.py\n..."  // generated patch (unified diff)
  },
  ...
}
```

--------------------------------

### DefaultAgent.execute_actions method

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/control_flow.md

Executes all actions in the response, formats observation messages, and adds them to the messages list.

```python
def execute_actions(self, message: dict) -> list[dict]:
    outputs = [self.env.execute(action) for action in message.get...
    return self.add_messages(*self.model.format_observation_messages(...))
```

--------------------------------

### GPT-5 reasoning effort configuration

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/models/quickstart.md

Configures reasoning effort and verbosity for GPT-5 mini. The drop_params flag is used to exclude unsupported parameters.

```yaml
model:
  model_name: "openai/gpt-5-mini"
  model_kwargs:
    drop_params: true
    reasoning_effort: "high"
    verbosity: "medium"
```

--------------------------------

### Configure default model name

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/global_configuration.md

Sets the default model name used by the agent.

```bash
# Default model name
# (default: not set)
MSWEA_MODEL_NAME="anthropic/claude-sonnet-4-5-20250929"
```

--------------------------------

### Configure LiteLLM Response model in agent config

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/reference/models/litellm_response_toolcall.md

Specify the model_class as litellm_response in your agent configuration file to enable the Responses API.

```yaml
model:
  model_class: "litellm_response"
  model_name: "openai/gpt-5.2"
  model_kwargs:
    drop_params: true
    reasoning:
      effort: "high"
```

--------------------------------

### DefaultAgent.run loop

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/control_flow.md

The main loop that calls step, catches InterruptAgentFlow exceptions, adds their messages, and breaks when a message with role='exit' is added.

```python
while True:
    try:
        self.step()
    except InterruptAgentFlow as e:
        self.add_messages(*e.messages)
    if self.messages[-1].get("role") == "exit":
        break
```

--------------------------------

### Agent protocol changes (removed attributes, run return type, save method)

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/v2_migration.md

Shows the removed attributes (model, env, messages), the changed run() return type (dict instead of tuple), and the new save() method for custom Agent classes.

```python
# Removed attributes (you don't need to implement these anymore, but you can)
model: Model
env: Environment
messages: list[dict]

# Changed return type
def run(self, task: str, **kwargs) -> dict: ...  # was: tuple[str, str]

# New method
def save(self, path: Path | None, *extra_dicts) -> dict: ...
```

--------------------------------

### Set litellm_model_registry in agent config

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/models/local_models.md

YAML snippet to add to the agent config file to set the path to the model registry JSON file.

```yaml
model:
  litellm_model_registry: "/path/to/model_registry.json"
  ...
...
```

--------------------------------

### Custom model registry JSON for litellm

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/models/local_models.md

JSON file following LiteLLM's format to override or add model cost and metadata. Includes entries for 'my-custom-model' and 'my-local-model' with max_tokens, costs, provider, and mode.

```json
{
  "my-custom-model": {
    "max_tokens": 4096,
    "input_cost_per_token": 0.0001,
    "output_cost_per_token": 0.0002,
    "litellm_provider": "openai",
    "mode": "chat"
  },
  "my-local-model": {
    "max_tokens": 8192,
    "input_cost_per_token": 0.0,
    "output_cost_per_token": 0.0,
    "litellm_provider": "ollama",
    "mode": "chat"
  }
}
```

--------------------------------

### Exit on submit command by subclassing the environment

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/cookbook.md

Subclass LocalEnvironment to raise a Submitted exception when the 'submit' command is issued. The agent is instantiated with the custom environment.

```python
from minisweagent.agents.default import DefaultAgent
from minisweagent.environments.local import LocalEnvironment
from minisweagent.exceptions import Submitted

class EnvironmentQuitsOnSubmit(LocalEnvironment):
    def execute(self, action: dict, cwd: str = "") -> dict:
        if action.get("command", "") == "submit":
            raise Submitted({
                "role": "exit",
                "content": "The agent has finished its task.",
                "extra": {"exit_status": "Submitted", "submission": ""},
            })
        return super().execute(action, cwd)

agent = DefaultAgent(
    LitellmModel(model_name=model_name),
    EnvironmentQuitsOnSubmit(),
)
```

--------------------------------

### Set global cost tracking via environment variable

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/models/local_models.md

Bash command to set the global MSWEA_COST_TRACKING environment variable to 'ignore_errors'. This affects all models.

```bash
export MSWEA_COST_TRACKING="ignore_errors"
```

--------------------------------

### LiteLLM Model Class

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/reference/models/litellm.md

The LiteLLM model class is defined in the file 'src/minisweagent/models/litellm_model.py'. The full source code is available on GitHub.

```python
class LiteLLMModel:
    # Full source code available at:
    # https://github.com/swe-agent/mini-swe-agent/blob/main/src/minisweagent/models/litellm_model.py
    pass
```

--------------------------------

### Unset a configuration key

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/advanced/global_configuration.md

Use the unset command to remove a specific configuration key.

```bash
mini-extra config unset KEY
# e.g.,
mini-extra config unset ANTHROPIC_API_KEY
```

--------------------------------

### Trajectory file structure (JSON)

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/usage/output_files.md

Shows the top-level structure of a trajectory file, including info, messages, and trajectory_format. The info object contains model_stats, config, mini_version, exit_status, and submission. Messages follow the OpenAI chat format with an extra field for metadata.

```json
{
  "info": {
    "model_stats": {
      "instance_cost": 0.05,  // total cost of API calls for this run
      "api_calls": 12  // number of API calls made
    },
    "config": {
      "agent": { ... },  // agent configuration
      "agent_type": "minisweagent.agents.default.DefaultAgent",
      "model": { ... },  // model configuration
      "model_type": "minisweagent.models.litellm_model.LitellmModel",
      "environment": { ... },  // environment configuration
      "environment_type": "minisweagent.environments.local.LocalEnvironment"
    },
    "mini_version": "2.0.0",  // version of mini-SWE-agent used
    "exit_status": "Submitted",  // final status (Submitted, LimitsExceeded, etc.)
    "submission": "..."  // final output/patch submitted by the agent (if any)
  },
  "messages": [  // full conversation history
    {"role": "system", "content": "..."},
    {"role": "user", "content": "..."},
    {"role": "assistant", "content": "..."},
    ...
  ],
  "trajectory_format": "mini-swe-agent-1.1"  // format version identifier
}
```

--------------------------------

### Ignore cost tracking errors in agent config

Source: https://github.com/swe-agent/mini-swe-agent/blob/main/docs/models/local_models.md

YAML snippet to add to the agent config file to ignore cost tracking errors. Sets cost_tracking to 'ignore_errors'.

```yaml
model:
  cost_tracking: "ignore_errors"
  ...
...
```
Tokens:
Raw