What problem does it solve? Research interview transcripts contain inconsistent spellings, misheard proper nouns, and undefined jargon that corrupt downstream correction and analysis. This Skill bootstraps and maintains a project dictionary in BigQuery so every future transcript is corrected against human-approved canonical terms. ## Core Features & Use Cases - Dictionary Bootstrapping: Builds the initial project dictionary from the first transcript, then reminds the orchestrator to re-run correction on that transcript once terms are approved. - Term Proposal with Evidence: Proposes clarification entries (term plus definition) and unification entries (one canonical term with variant spellings), each backed by quoted evidence lines and a calibrated confidence score. - Human-Gated Approval: Writes only proposed rows and never sets active status, ensuring a human decides before any term enters the dictionary. - Dictionary Serving: Returns only active terms to other agents such as Scribe and Tagger, never leaking unapproved proposals. - Use Case: After ingesting the first client interview, the agent proposes canonical forms like Acme Corp with variants ACME, Acme, and A.C.M.E., each with transcript quotes, so a human can approve them and future transcripts are auto-corrected. ## Quick Start Ask the agent to scan the first ingested transcript and propose canonical dictionary terms with evidence for human review.
