What problem does it solve? Answering "what would I likely choose" or drafting outward-facing text in your own voice requires consulting a personal profile without treating past behavior as identity, leaking private evidence, or fabricating first-person claims. This Skill keeps profile facts, predictions, advice, reflections, and drafted language strictly distinct while consulting the profile read-only. ## Core Features & Use Cases - Four labeled modes: predict (likelihood with confidence and counterevidence), advise (profile alignment separated from external merits and recommendation), reflect (confirmed patterns and tensions without diagnosis), and draft (outward-safe text using only confirmed outward-safe assertions). - Strict evidence governance: validates the sd-personal-profile/v1 schema, selects exactly one audience/channel overlay, classifies request-scoped current context separately from durable profile evidence, and enforces visibility rules so private-only entries never reach outward drafts. - Use Case: Ask "Would I likely accept this conference speaking invitation?" and receive a labeled prediction citing relevant confirmed assertion IDs, dates, strongest counterevidence, and a qualitative confidence level, with an explicit statement that prediction is not consent or commitment. ## Quick Start Ask the assistant to use sd-ask-me with mode=predict to answer what you would likely choose between two job offers based on your approved profile.
