wehub-resource-sync
49 lines
1.3 KiB
ReStructuredText
49 lines
1.3 KiB
ReStructuredText
Custom Experiments
|
|
==================
|
|
|
|
The primary custom-experiment path is the setup wizard.
|
|
|
|
Browser flow
|
|
------------
|
|
|
|
Run:
|
|
|
|
.. code-block:: bash
|
|
|
|
./scripts/god.sh configure
|
|
|
|
Then choose **Create Custom Experiment**. The wizard asks for scenario context, generates a draft, lets you edit agents and steps, and publishes the result as the current experiment.
|
|
|
|
Published shape
|
|
---------------
|
|
|
|
Custom experiments are written under:
|
|
|
|
.. code-block:: text
|
|
|
|
agentsociety/quick_experiments/hypothesis_<slug>/experiment_1/
|
|
|
|
The important files are:
|
|
|
|
``init/init_config.json``
|
|
World context, map module, initial locations, agent profiles, enabled skills, and runtime-facing config.
|
|
|
|
``init/steps.yaml``
|
|
Start timestamp, step count, and tick duration.
|
|
|
|
``README.md`` and ``README.zh-CN.md``
|
|
Human explanation for the experiment.
|
|
|
|
``run.sh``
|
|
Optional foreground runner for debugging.
|
|
|
|
Current experiment pointer
|
|
--------------------------
|
|
|
|
When you click **Save and Launch**, setup writes ``.god/current_experiment.json``. The shell script then starts that experiment without changing ``.env``.
|
|
|
|
Manual editing
|
|
--------------
|
|
|
|
You can edit experiment files directly, but the setup wizard is safer for first drafts because it keeps IDs, map selection, and runtime config aligned.
|