Skip to main content
Put and read the system prompts to score models under, each by its pr_… id Every ambral command also takes the global flags.

ambral prompts put

Put a prompt and print its id, derived from its exact text: the same text always has the same id, and any edit, whitespace included, makes another. ambral runs start --subjects <model>@<id> asks a model under it: it replaces the system prompt each trace opens with, or is added in front of one with none Examples

ambral prompts get

Print a prompt’s text, exactly as it was put or as a taskset’s traces sent it, as a text file holds it: ambral prompts put reads that file back under the same id. With --json, its id and exact text Examples