correct_bars
ChatGPTUse this after read_pdf for a misread bar the user describes, or one read_pdf flagged or you can see is misread when you are confident what it should be. Pass its score_id and replacement_bars in N1 format, replacing every voice of each listed bar and including every staff. Example: N1 div=24 ts=4/4 ks=0 m3 len=96 1.1: 24 C4 D4 48 E4 div is units per quarter note; m3 is printed bar 3; len is its length in those units. Each staff.voice line uses numbers to set the duration for following tokens, + for chords, r for rests, g for grace notes (u for slashed ones) and ~ for ties to the next note; pitches are sounding pitches (e.g. F#4, Bb3). Every voice must explicitly fill the existing bar, including rests. Header key/meter do not change the score. Use part (a part id or name) in a score with multiple parts; staff numbers count within that part. Returns a new score_id, PNGs of only changed bars and a note per bar; runs no model. If you proposed the correction, say it is your suggestion. Show the redrawn bars and note; ask the user to confirm against their copy. Only after confirmation analyse with draft_analysis using the new score_id. If they decline, keep using the original score_id.
draft_analysis
ChatGPTUse this when the user wants a harmonic analysis of a MusicXML score (.mxl or .musicxml): Roman numerals, cadences, phrases, non-chord tones. Returns a neural model's draft as a table for you to review and correct, plus a score_id. Follow it with render_analysis. Give the score in exactly one way. score_file: a file the user attached to the chat. score_id: the id returned by an earlier draft_analysis. Never encode, paste or retype a file's contents. If the user has not attached a score, ask them to attach their MusicXML file (.mxl or .musicxml) to the chat.
draft_from_notes
ChatGPTUse this when a skill has produced a compact note list of the user's score (it starts with a line like 'N1 div=48 ts=12/8 ks=3f') because the score file itself cannot be sent. Pass that text unchanged as notes. Returns a neural model's draft analysis (chords, cadences, phrase ends, possible non-chord tones) as a table for you to review and correct before you annotate the score. Without such a note list you cannot analyse the score: tell the user that in Claude this needs the Sheet Music Analysis plugin (from Claude's directory) and Code execution and file creation turned on in Settings > Capabilities.
rate_analysis
ChatGPTRecords the user's thumbs up or down from the analysis card.
read_pdf
ChatGPTUse this when the user attaches a PDF of printed sheet music (born-digital or scanned) and wants it analysed. Reads the music from the PDF and shows the user the engraved result in a card, so they can check it against their copy. Returns a score_id. Do not analyse yet: wait until the user says to go ahead, then call draft_analysis with that score_id. PDFs of more than 12 pages are refused. Slurs are left out unless keep_slurs is true; set it only when the user asks for them.
render_analysis
ChatGPTUse this after draft_analysis, once you have reviewed the draft, to draw your analysis on the engraved score. Takes the score_id and your annotation list and returns the page as an image plus PNG and PDF links, shown to the user in a card. If a position does not exist in the score, the error lists each problem so you can fix the list and call again. A page shows at most 32 bars: for a longer piece, pass measure_range (e.g. 1-16) and draw it passage by passage. view is one of all, harmony, voice_leading, form.
report_problem
ChatGPTRecords that the user sent us the problem shown in the card, by its reference.
send_suggestion
ChatGPTRecords a suggestion the user typed into the card.
choose_score
Claudecorrect_bars
ClaudeUse this after read_pdf for a misread bar the user describes, or one read_pdf flagged or you can see is misread when you are confident what it should be. Pass its score_id and replacement_bars in N1 format, replacing every voice of each listed bar and including every staff. Example: N1 div=24 ts=4/4 ks=0 m3 len=96 1.1: 24 C4 D4 48 E4 div is units per quarter note; m3 is printed bar 3; len is its length in those units. Each staff.voice line uses numbers to set the duration for following tokens, + for chords, r for rests, g for grace notes (u for slashed ones) and ~ for ties to the next note; pitches are sounding pitches (e.g. F#4, Bb3). Every voice must explicitly fill the existing bar, including rests. Header key/meter do not change the score. Use part (a part id or name) in a score with multiple parts; staff numbers count within that part. Returns a new score_id, PNGs of only changed bars and a note per bar; runs no model. If you proposed the correction, say it is your suggestion. Show the redrawn bars and note; ask the user to confirm against their copy. Only after confirmation analyse with draft_analysis using the new score_id. If they decline, keep using the original score_id.
draft_analysis
ClaudeUse this when the user wants a harmonic analysis of a MusicXML score (.mxl or .musicxml): Roman numerals, cadences, phrases, non-chord tones. Returns a neural model's draft as a table for you to review and correct, plus a score_id. Follow it with render_analysis. If the user wants their own analysis checked, pass their Roman numerals as student_analysis, as they typed them ("m1 I, m2 V6, m3 beat 3 ii65 ..."), or the word "score" when the numerals are written into their score file; the draft then lists, beat by beat, where they agree with it. If the result is a question, ask the user that question and call again with their answer. Give the score in exactly one way. score_file: a file the user attached to the chat. score_id: the id returned by an earlier draft_analysis, read_pdf or correct_bars. Never encode, paste or retype a file's contents. If the user has not attached a score, ask them to attach their MusicXML file (.mxl or .musicxml) to the chat.
draft_from_notes
ClaudeDrafts a harmonic analysis from a compact note list of a score: text starting with a line like 'N1 div=48 ts=12/8 ks=3f', which the Sheet Music Analysis plugin's skill makes from the user's score file, passed unchanged as notes. Returns a neural model's draft as a table: chords as Roman numerals, cadences, phrase ends and possible non-chord tones, with part-writing findings where they apply. With student_analysis, the user's own Roman numerals as the skill's script prints them, it also lists beat by beat where they agree with the draft. Without a note list, or when the user's numerals are ambiguous, the result is an error that says what is missing.
fetch_score
Claudeplay
Clauderead_fetched_pdf
Clauderead_pdf
ClaudeUse this when the user attaches a PDF of printed sheet music (born-digital or scanned) and wants it analysed. Reads the music from the PDF and shows the user the engraved result in a card, so they can check it against their copy. Returns a score_id. Do not analyse yet: wait until the user says to go ahead, then call draft_analysis with that score_id. PDFs of more than 12 pages are refused. Slurs are left out unless keep_slurs is true; set it only when the user asks for them.
render_analysis
ClaudeUse this after draft_analysis, once you have reviewed the draft, to draw your analysis on the engraved score. Takes the score_id and your annotation list and returns the page as an image plus PNG, PDF and MusicXML links, shown to the user in a card. If a position does not exist in the score, the error lists each problem so you can fix the list and call again. A page shows at most 32 bars: for a longer piece, pass measure_range (e.g. 1-16) and draw it passage by passage. view is one of all, harmony, voice_leading, form. Set annotations.style to abrsm (letter inversions, Ib, Vc) when the user mentions ABRSM, figured_bass when they ask for figured bass, otherwise leave it out. Offer the MusicXML link once per conversation, not after every render: it downloads a file for MuseScore, Sibelius or Dorico; the result also lists a worksheet PDF, MIDI and RomanText to mention once. piano=true draws the same annotation list (written for the original score) on a two-staff piano reduction with instrument cues. For a score with six or more instruments (an orchestra, band or large ensemble; not a quartet), draw the original first, then offer the user the piano reduction once; call again with piano=true only if they accept.
send_suggestion
Claude