{# T55 sandbox editor — terminal-brutalist re-skin. #} {# system + user prompt textareas + model dropdown + run button. #} {% extends "admin_layout.html" %} {% block page_title %}admin :: sandbox :: {{ prompt_name }}{% endblock %} {% block breadcrumb %} admin/ experiments/ {{ prompt_name }} {% endblock %} {% block content %}
08.1 — prompt sandbox

Tune the prompt. {{ prompt_name }}

Edit either prompt, pick a model, then [run experiment]. This never overwrites the original call — a fresh llm_experiments row is created. Cost is billed to your admin account (see the daily cap on /admin/sandbox/list).

{# ===================== Source meta ===================== #}
source
source call{{ source_call_id }}
{% if scan_id %}
scan{{ scan_id }}
{% endif %}
prompt{{ prompt_name }}
{# T36.4 — empty-state notice when the source dump is missing/malformed. #} {% if not source_dump_available %}
[!] source dump unavailable

The dump format predates the current schema, or the S3 object was deleted. You can still type fresh prompts manually below.

{% endif %} {# ===================== Form ===================== #}
editor
{# T36.3 — original response as a chat bubble. #}
original response
{% with chat_view_id='original-response-' ~ source_call_id, inline_dump=original_dump_json %} {% include 'partials/llm_chat_view.html' %} {% endwith %}
all experiments · admin dashboard {% if scan_id %} · source debug page {% endif %}
{% endblock %} {% block scripts %} {% endblock %}