What problem does it solve? Tagging research transcripts at scale is error-prone: untagged lines and never-read lines look identical, so partial tagging runs silently appear complete. This Skill separates judgement from bookkeeping — the model decides which tags each line earns, while a deterministic script serves batches, validates every tag against the active library, and proves the batches tiled the entire conversation. ## Core Features & Use Cases - Batch-driven tagging loop: A three-command cycle (next-batch, write-tags, status) serves 50-line ranges with surrounding context until coverage is complete. - Validated writes: Every tag is checked against the active tag library, per-tag confidence thresholds, batch line ranges, and non-empty justifications before any row is written to BigQuery. - Coverage proof: The status command derives completeness from recorded batch ranges and reports lines considered but untagged, distinguishing real zero-tag results from unread lines. - Use Case: After Scribe ingests a user-research interview transcript, run this Skill to apply the shared tag dictionary to every line, producing auditable tag rows with confidence scores and justifications. ## Quick Start Ask the agent to tag the latest ingested conversation by running the tagger batch loop until the status command reports complete coverage.
