delete_library
ChatGPTPermanently delete this anonymous library and its saved lessons, transcripts, attempts, drafts, preferences, and preparations. Available only to the app after the user confirms deletion. Pass the current exact library ID in both libraryId and confirmation; this is not an authenticated identity.
get_attempt
ChatGPTRead the canonical answer, lessonVersion, status, and saved feedback for this exact attempt in the anonymous library namespace specified by libraryId. When grading, use this answer (never a draft or another attempt), then call reveal_segment with this attempt’s lessonId and segmentId and save_feedback with this attempt’s lessonVersion. Do not reopen the Listen Lab UI. Pass the exact listenLabLibraryId from current model context as libraryId; never invent or derive it from an email.
get_lesson
ChatGPTGet lesson metadata and clip cue boundaries without revealing transcript text. Pass the exact listenLabLibraryId from current model context as libraryId; never invent or derive it from an email.
get_lesson_progress
ChatGPTRecover current-version answer drafts and the latest attempt for each current lesson segment. Returns no transcript text and is available only to the app. Pass the exact listenLabLibraryId from current model context as libraryId; never invent or derive it from an email.
get_user_state
ChatGPTRead persisted interface language, explanation language, clip length preference, and active lesson and clip. Pass the exact listenLabLibraryId from current model context as libraryId; never invent or derive it from an email.
list_lessons
ChatGPTList lessons in the anonymous library namespace. Pass the exact listenLabLibraryId from current model context as libraryId; never invent or derive it from an email.
list_transcript_cues
ChatGPTRead an ordered page of transcript cues for a preparation. Continue using nextCursor until it is null before making a complete lesson plan. Pass the exact listenLabLibraryId from current model context as libraryId; never invent or derive it from an email.
open_listen_lab
ChatGPTUse only when the user explicitly asks to open Listen Lab, browse the lesson library, or navigate to a lesson. On the first launch only, omit libraryId when current model context has no listenLabLibraryId; this returns an empty home. Otherwise pass the exact ID from current model context. Never invent a library ID. Do not call this tool when checking, submitting, or resuming an answer attempt; it reloads the app. For grading, use get_attempt, then reveal_segment, then save_feedback without reopening the UI. Treat the lesson and excerpt currently shown in the app as authoritative; do not announce or summarize the selected lesson unless the user asks.
prepare_video
ChatGPTFetch the original-language caption track for a YouTube video and create a transcript preparation. This returns the first cue page only; use list_transcript_cues with increasing numeric cursors until nextCursor is null before selecting cue ranges. Never invent cue IDs or timestamps. Pass the exact listenLabLibraryId from current model context as libraryId; never invent or derive it from an email.
resegment_lesson
ChatGPTReplace the clip boundaries for an existing lesson using its saved transcript cues. Include every cue currently covered by the lesson exactly once, in order; do not add or drop cues. Send expectedVersion from the current lesson; on a stale version, retry safely only when the same ordered cue ranges are already saved as the next version. The server increments the version atomically, keeps old segment rows and attempts, and never fetches the video again. When clipLengthPreference is short, a multi-cue range over 15 seconds is rejected if shorter cue-boundary splits exist; a naturally long single cue is allowed. Other preferences do not apply this short-clip limit. Pass the exact listenLabLibraryId from current model context as libraryId; never invent or derive it from an email.
reveal_segment
ChatGPTExplicitly reveal the exact stored transcript cues for one saved clip. Pass the exact listenLabLibraryId from current model context as libraryId; never invent or derive it from an email.
save_answer_draft
ChatGPTPersist the learner’s current answer in the anonymous library namespace specified by libraryId for recovery after an app reload. Include the current lesson version and a revision greater than the last saved revision. Empty answers are valid drafts. The app-only progress tool does not return draft text to the model. A library ID is not authenticated identity or a privacy boundary. Pass the exact listenLabLibraryId from current model context as libraryId; never invent or derive it from an email.
save_feedback
ChatGPTGrade only the exact attempt returned by get_attempt for the supplied attemptId in the specified anonymous library namespace; use that attempt’s answer and lessonVersion, and do not substitute a draft or another submission. Read its exact clip transcript with reveal_segment using the lessonId and segmentId from that attempt, then save feedback with the same lessonVersion. Corrections reference ordered cue IDs within that attempt’s clip; do not include original transcript text. Transcript text is not returned until reveal_segment is explicitly requested. Do not reopen the Listen Lab UI to grade. Pass the exact listenLabLibraryId from current model context as libraryId; never invent or derive it from an email.
save_lesson_plan
ChatGPTSave ordered, non-overlapping inclusive ranges from the prepared transcript, using only real startCueId/endCueId values. The server derives all clip timings from those cues. Save one plan per preparation. Pass the exact listenLabLibraryId from current model context as libraryId; never invent or derive it from an email.
submit_attempt
ChatGPTSave a learner’s answer as a pending attempt. Use an idempotencyKey stable for retries of the same user submission. Pass the exact listenLabLibraryId from current model context as libraryId; never invent or derive it from an email.
update_user_state
ChatGPTPersist language preferences, clip length preference, and active lesson and clip within the anonymous library namespace specified by libraryId, across client contexts that use that same ID. The ID is not an authenticated identity. Pass the exact listenLabLibraryId from current model context as libraryId; never invent or derive it from an email.