Every skill page carries a Test against live API panel: a conversation that actually executes the skill's operations against the real application. Ask in plain words — every call the skill makes shows up in the transcript as evidence you can expand.
A fresh session greets you with what the skill can do and offers starter chips built from its top read operations — these don't change anything. If the skill is mounted in playbooks, you can also start from a playbook's instructions.
Type your own question in the composer (Enter sends, Shift+Enter adds a line). New session clears the transcript and starts over.
Each API call appears as a collapsible evidence block showing the method, URL, HTTP status, and duration. Expand it for the parameters that were sent (with where each value came from) and the response body. The assistant's reply is grounded in what the API actually returned — the raw data is right there to check.
The Allow write calls switch is off by default: the skill can only execute read operations, and anything marked WRITE in the manifest is refused. Flip the switch and a warning ribbon appears — "Write calls enabled — POST and DELETE will hit the real API." The switch is per session and resets with it.
The Session params pill in the panel header lists the skill's variables. Each shows its saved default; edit one and the new value applies to this session only — your saved defaults are untouched. A counter warns when required variables have no value.
The chip in the panel header shows your workspace balance and the price per turn — 1 credit for a Light skill (up to 20 operations), 3 for Standard (21–50), 6 for Heavy (51+).
See Credits & plans for the full pricing model.
