القائمة
Use when composing music with the music-server MCP tools. Covers note syntax, tool workflow, and composition patterns.
التثبيت باستخدام Codex أو Claude انسخ هذا Prompt والصقه في Codex أو Claude أو مساعد آخر ليراجع صفحة Skill ويثبّتها لك.
استنادا إلى تصنيف SOC المهني
| name | music-composition |
| description | Use when composing music with the music-server MCP tools. Covers note syntax, tool workflow, and composition patterns. |
You have 47 MCP tools for building compositions from scratch. This skill covers the syntax, tool flow, and patterns.
Before doing anything else, load all music-server schemas in one batch ToolSearch call:
ToolSearch: "mcp music-server" (max_results=50)
Do this once at the start of the session. MCP tool schemas are deferred — calling any tool without loading its schema first causes an error and forces a mid-composition round-trip.
Notes are whitespace-separated tokens: pitch/duration
Pitch: standard name + optional accidental + octave: C4, F#3, Bb5, Eb4
Rests: r/q, r/h, r/e etc.
Chords: angle brackets around space-separated pitches, followed by /duration:
<C4 E4 G4>/q → C major triad, quarter note
<Eb3 G3 Bb3>/h → Eb major triad, half note
<D3 F3 A3 C4>/q → Dm7 chord, quarter note
Use chords for pad stabs, gospel organ, jazz voicings — any simultaneous pitches.
Duration codes:
| Code | Value |
|---|---|
w | whole |
h | half |
q | quarter |
e | eighth |
s | sixteenth |
dh | dotted half |
dq | dotted quarter |
de | dotted eighth |
h3 | half-note triplet (= 1/3 whole) |
q3 | quarter-note triplet (= 1/6 whole) |
e3 | eighth-note triplet (= 1/12 whole) |
s3 | sixteenth-note triplet (= 1/24 whole) |
Ties — append ~tie to hold a note into the next note of the same pitch (extends sound across bar lines):
C4/h~tie C4/q → C4 sounds for a dotted half total
G4/q~tie G4/e → G4 quarter + eighth tied together
Ties are rendered as ~ in LilyPond and extend NOTE_OFF in MIDI. Use for notes that cross bar lines or need durations not expressible as a single rhythmic value.
Per-note velocity — append ~vel:N (0–127) to override the voice-level dynamic for a single note:
C4/q~vel:100 D4/q E4/q~vel:40 # loud C, default D, quiet E
G4/h~vel:120~tie G4/q # very loud tied note
Useful for shaping phrases: push peak notes, pull back resolutions, accent approach tones. Takes precedence over voice.set_dynamics.
Articulations — append ~code after the duration:
| Suffix | Articulation | LilyPond mark | MIDI effect |
|---|---|---|---|
~stac | staccato | -. | 50% duration |
~acc | accent | -> | normal duration |
~ten | tenuto | -- | full duration |
~marc | marcato | -^ | 75% duration |
~leg | legato | (none) | full duration |
C4/q~stac D4/q E4/h # staccato C, normal D and E
<C4 E4 G4>/q~acc # accented chord stab
Use voice.set_articulation to apply an articulation to an entire voice at once.
Ornaments — append ~code after the duration (combinable with articulations and ~vel:N):
| Suffix | Ornament | LilyPond glyph | MIDI approximation |
|---|---|---|---|
~roll | roll (Irish/folk) | \trill | main–upper–main–lower–main |
~trill | trill | \trill | rapid main–upper alternation |
~mord | mordent (lower) | \mordent | main–lower–main |
~prall | pralltriller | \prall | main–upper–main |
A4/e~roll # rolled A eighth — classic Irish jig ornament
B4/q~trill # trilled B quarter
G4/e~mord # mordent on G eighth
MIDI uses semitone neighbors as pitch approximation (upper = +2, lower = −1). For accurate neighbor pitches, use explicit grace notes instead.
Grace notes — prefix {pitch ...} before the main note token. Pitches are space-separated; use standard pitch notation:
{G4}A4/q # single grace note G before A quarter
{G4 A4}B4/e # two grace notes before B eighth
Grace notes steal time from the front of the main note's sounding duration (~32nd note each). They appear as \grace { ... } in LilyPond.
Control Change (CC) events — point-in-time MIDI controller values placed inline in a voice sequence. Zero duration; do not advance time. MIDI only — invisible to LilyPond/MusicXML.
cc:pan:64 → center pan (CC 10, value 64)
cc:pan:20 → hard left
cc:pan:110 → hard right
cc:expr:100 → high expression (CC 11)
cc:expr:40 → low expression — fade down a phrase
cc:reverb:60 → medium reverb send (CC 91)
cc:mod:80 → modulation / vibrato depth (CC 1)
cc:sustain:127 → sustain pedal down (CC 64)
cc:sustain:0 → sustain pedal up
cc:11:80 → numeric form — same as cc:expr:80
Named aliases: mod (1), vol (7), pan (10), expr (11), sustain (64), reverb (91), chorus (93).
Expression CC (11) is how you shape dynamics on strings and brass — velocity sets the attack, expr shapes the phrase arc. Pan places instruments in the stereo field. Place CC events immediately before the note they should affect:
voice.create "strings" "cc:pan:30 cc:expr:60 C4/h cc:expr:90 E4/h cc:expr:70 G4/w"
Program Change events — mid-score instrument change within a voice. Zero duration. Use for switching patches (e.g. arco to pizzicato).
pc:45 → switch to Pizzicato Strings (0-indexed GM program; GM tables list this as #46)
pc:40 → switch back to Violin (0-indexed GM program; GM tables list this as #41)
voice.create "strings" "C4/h D4/h pc:45 E4/q F4/q G4/h"
Examples:
C4/q D4/q E4/q F4/q # four quarter notes
E4/e F#4/e G4/q r/q # eighths, quarter, rest
C4/dq D4/e E4/h # dotted quarter + eighth + half
<Eb3 G3 Bb3>/q <Ab3 C4 Eb4>/q # two chord stabs
C4/q~stac D4/q~stac r/e E4/e F4/q # staccato melody
score.set_metadata — set title, tempo (BPM), time signature ("4/4", "3/4")score.set_swing — enable swing quantization: ratio="2/3" (standard jazz), "3/5" (light), "0" to disablescore.set_tempo_change — gradual ritardando or accelerando across a bar span: start_bar, end_bar, to_bpm, optional curve="linear" (default) or "exponential" (more natural for rit.). Affects both MIDI playback and LilyPond notation.score.describe — inspect full composition state (voices, dynamics, articulations, instruments)score.assign_instrument — assign a voice to a GM instrument (see list below)score.clear — wipe everything and start freshvoice.create — create a named voice from a note sequencevoice.append — append more notes to an existing voicevoice.list — list all voices with bar counts, event counts, and durationvoice.measure_count — return the number of complete bars in a single voicevoice.set_dynamics — set volume for a voice: ppp pp p mp mf f ff fff (default: mf)voice.set_articulation — set articulation for a voice: staccato accent tenuto marcato legato. Optional from_bar / to_bar to scope to a bar range (e.g. make only the final phrase tenuto).voice.add_expression_curve — add a gradual CC 11 expression curve to a voice. Parameters: voice, start_bar, end_bar, from_value (0–127), to_value (0–127), curve (linear or exponential). Emits one CC 11 event per bar between start_bar and end_bar, interpolated from from_value to to_value. Use for crescendos, decrescendos, and phrase swells. Multiple curves on the same voice are additive. Example: voice.add_expression_curve voice=melody start_bar=9 end_bar=16 from_value=30 to_value=110 curve=exponential.voice.from_motif — Advanced: create a voice from a saved motif, optionally transformedvoice.delete — remove a voice from the compositionvoice.trim — truncate a voice to N bars, discarding everything aftervoice.set_bar — replace one bar's content: bar=N notes="C4/q D4/q E4/q F4/q". Use query.voice to inspect before editing.voice.replace_range — replace bars from_bar–to_bar (inclusive) with a new note sequence. The replacement doesn't have to fill the same number of bars.voice.replace_note — surgical fix: find the first note in bar N whose pitch matches old, replace it with new. Both old and new are full note tokens including duration, e.g. old="C4/q" new="D4/q". Use query.voice to inspect bar content first.query.voice — show the note sequence of a voicequery.motif — show the note sequence of a motiftransform.transpose — transpose by interval (e.g. "P5", "m3", "M2", "P8"), direction "up"/"down"transform.invert — Advanced: mirror pitches around an axis pitch (e.g. "C4")transform.retrograde — Advanced: reverse temporal ordertransform.augment — Advanced: scale all durations by a factor ("2/1"=double, "1/2"=halve)motif.save — save a slice of a voice as a named motif (start/end are 0-based note indices)voice.concat — concatenate multiple voices sequentially (voice_names: comma list, target_voice)voice.repeat — repeat a voice N times — great for ostinatos and drum loopsvoice.slice — Advanced: extract measures startMeasure–endMeasure (exclusive) from a voicevoice.pad_to_measure — Advanced: prepend rests so the voice starts at the given measure numberharmony.set_key — set the composition key, e.g. "F major", "D minor", "Bb major"harmony.chord_progression — set a Roman-numeral progression, e.g. "I IV V I", "ii V I", "I V vi IV"harmony.set_bars — declare per-bar chord changes as concrete chord symbols: "1:Cm7 2:F7 3:Bbmaj7 4:Eb"harmony.harmonize — generate a chord-root voice from the current key + progression (target_voice, optional octave)harmony.walking_bass — generate a walking bass line over chords; adapts to time signature: 4/4 = root-fifth-third-approach, 3/4 = root-fifth-fifth, 6/8 = root-fifth (dotted-quarters), 2/4 = root-approach. Uses harmony.set_bars if set, else falls back to progression. Optional: velocity (ppp–fff, default mf), approach (chromatic [default, 1–2 semitones below], diatonic [scale step below], none [use chord root])harmony.comp — generate a comping (chord accompaniment) voice; adapts to time signature. Style options: quarter_stabs (default), on_beat, eighth_pump, shell_voicings, charleston (4/4 only). Optional: velocity (ppp–fff, default mf)harmony.suggest_harmony — suggest chord harmonizations measure-by-measure for a given voice (advisory)harmony.detect_key — Advanced: run pitch-class analysis on a voice and report the most likely keyharmony.diatonic_transpose — transpose a voice by N diatonic scale steps within the current keydrums.preset — load a named drum pattern for N bars; creates drums_kick, drums_snare, etc. voices on channel 9
house_4on4 — four-on-the-floor kick every beat + clap on 2 & 4 + closed 8th hi-hats; the EDM/dance foundation. Use at 120–130 BPM.rock_8th — driving 8th hi-hat, snare on 2 & 4, kick on 1 & 3; heavier and faster than rock_basic. Use at 130–180 BPM.rock_basic — standard rock backbeat: quarter hi-hat, snare on 2 & 4, kick on 1 & 3; backbone of rock and pop-rock. Use at 100–160 BPM.swing — jazz ride cymbal pattern, hi-hat on 2 & 4, snare on 2 & 4; pair with score.set_swing "2/3" for authentic jazz feel. Use at 120–220 BPM.waltz — simple 3/4: kick on beat 1, snare on beats 2 & 3; works for classical, folk, or pop waltzes. Use at 100–160 BPM.waltz_jazz (3/4) — kick + hi-hat on beat 1, soft brushed snare (MP) on beats 2 & 3; more idiomatic than waltz for jazz waltz contexts. Pair with score.set_swing "2/3".bossa_nova — defining rhythm is the tresillo clave (3+3+2 eighths, i.e., dq+dq+q); creates drums_kick, drums_rim, drums_hihat (eighth hi-hat), drums_clave. Do NOT add swing — the clave replaces swing feel.afrohouse — syncopated (not 4-on-the-floor) kick + open hi-hat on upbeats + hand percussion; creates drums_kick (beat 1, and-of-2, beat 3), drums_clap (2&4), drums_hihat (8ths + open on upbeats), drums_congas (hi/low alternating), drums_maracas (running eighths, velocity P). After loading: set drums_maracas to p or pp and drums_congas to mp.rules.check — run voice-leading and meter rules; percussion voices (channel 9) are automatically skippedrules.check_range — check all notes in a voice against a named instrument's written rangeinstrument.info — get written range, comfortable range, MIDI program, clef, and articulations for a named instrumentform.create_section — capture the current voices (or a subset) as a named section. measures is optional — inferred from the longest voice if omitted; a warning is added if explicit measures differs from the inferred count.form.repeat_section — append a repeat of a previously defined section (optionally with a new label)form.set_ending — attach a 1st/2nd volta ending to a section: section (the repeated section), pass (1 or 2), ending_section (a separately created section with the alternate tail bars). LilyPond renders \repeat volta + \alternative blocks; MIDI plays the correct ending per pass.form.build — assemble sections into a full score; replaces per-section voices with the full assembled versionsform.describe — show the current formal plan (e.g. "A(8) → B(8) → A(8), total: 24 measures")export.all — preferred — creates a folder with .mid, .ly, .json, and PDF (if LilyPond installed). The server auto-prepends a sequential number to the folder name (e.g. folder="sunlit_yard" → 25_sunlit_yard). Do NOT include a number yourself — it will double-prefix (e.g. 25_25_sunlit_yard).export.midi — write a .mid file to the current directoryexport.lilypond — write a .ly file (and engrave PDF if LilyPond is installed)export.musicxml — write a MusicXML 4.0 file (.musicxml). Opens in MuseScore, Dorico, Sibelius, Finale, and most DAWs.score.save — save the full composition state to a JSON snapshot under generated_tracks/ (voices, motifs, metadata, key, swing, bar chords, tempo changes). Use to checkpoint work in progress.score.load — restore a previously saved snapshot, replacing all current state.score.load_midi — import a MIDI file as voices (voice-0, voice-1, …) and set tempo from the file. Existing voices are preserved; call score.clear first for a clean slate. Black keys come back as sharps — re-spell with transform.transpose if needed. path is absolute or relative to the server working directory.score.load_abc — parse an ABC notation string into one or more voices. Headers T: (title), L: (default note length), M: (meter), Q: (tempo), K: (key) are all read. Multi-voice files (V: body headers) produce one voice per V: section, named by the voice ID (e.g. V:soprano → voice soprano, V:1 → voice 1). Single-voice files produce one voice named after the T: field (lowercased, spaces → underscores). Key-signature accidentals apply automatically and persist within each measure; explicit accidentals (^, _, =) override for the remainder of the bar. Octave markers ' (raise) and , (lower) are supported. Tied notes (-) are merged into one longer note. Chord events ([CEG]) are imported as simultaneous pitches. Tuplets ((3, (p:q:r) are duration-scaled correctly (e.g. triplet eighth = 1/12 whole). Use to load folk tunes from The Session (thesession.org) or any ABC corpus. Existing voices are preserved; call score.clear first for a clean slate.Keyboards: piano, bright_piano, honky_tonk (detuned upright), electric_piano (Rhodes), rhodes, harpsichord, organ, accordion
Tuned percussion: vibraphone, marimba
Guitar / bass: guitar, electric_guitar, electric_bass, synth_bass
Strings: violin, viola, cello, strings, synth_strings
Choir: choir
Brass / winds: trumpet, trombone, tuba, french_horn, brass, saxophone, clarinet, flute, oboe, bassoon
Synth: synth_lead (sawtooth — classic house/trance lead), synth_pad (warm pad — chord stabs)
Drums: drums — automatically assigned to MIDI channel 9 (GM percussion). Notes in the voice are treated as GM drum sound numbers. For the full note-to-sound map, read skills/music-composition/drum-note-map.md. Common notes: C2 kick, D2 snare, F#2 closed hi-hat, Bb2 open hi-hat, Eb2 clap, Eb3 ride.
1. score.set_metadata — title, tempo, time sig
2. harmony.set_key — optional; required for harmony.* and diatonic_transpose
3. drums.preset — optional; quick drum pattern for N bars
4. voice.create — one call per voice, with full note sequence
voice.append — add more bars if needed
voice.set_dynamics — set volume per voice (f for lead, mp for pads, etc.)
voice.set_articulation — set articulation per voice (staccato for stabs, etc.)
voice.add_expression_curve — add a CC 11 crescendo/decrescendo curve (from_value→to_value over bar range)
5. transform.transpose — derive harmony voices (parallel thirds, octave doublings)
voice.repeat — extend loops to full length
voice.concat — join sections manually if not using form tools
6. harmony.harmonize — generate an accompaniment voice (needs harmony.set_key first)
7. rules.check — validate voice leading and meter (percussion skipped automatically)
8. form.create_section — snapshot current voices as a named section (A, B, etc.)
form.repeat_section — plan repeats (ABA, AABA, etc.)
form.build — assemble sections; voices replaced with full-length assembled versions
9. score.assign_instrument — one call per voice (instrument assignments survive form.build, so either order works)
10. score.describe — sanity check before export
11. export.all — or export.midi / export.lilypond
House track with chord stabs:
harmony.set_key "Eb major"
voice.create "chords" "<Eb3 G3 Bb3>/q r/q <Ab3 C4 Eb4>/q r/q"
voice.set_dynamics "chords" "mp"
voice.set_articulation "chords" "staccato"
score.assign_instrument "chords" "synth_pad"
Walking bass line (manual):
harmony.set_key "F major"
voice.create "bass" "F2/q A2/q C3/q E3/q | D2/q F2/q A2/q C3/q"
voice.set_dynamics "bass" "f"
score.assign_instrument "bass" "electric_bass"
Diatonic harmony in thirds:
harmony.set_key "G major"
transform.transpose "melody" "M3" "up" target_voice="melody_thirds"
# Now melody_thirds is a parallel major-third above melody (chromatic, not diatonic)
# For diatonic thirds use harmony.diatonic_transpose "melody" steps=2
Structural form (ABA):
# Build A section voices, then:
form.create_section "A" measures=8
# Build B section voices (can be completely different), then:
form.create_section "B" measures=8
form.repeat_section "A"
form.build # assembles A(8) + B(8) + A(8) = 24 measures
# After form.build, all voices are replaced with full 24-bar assembled versions
Volta endings (1st/2nd ending brackets):
# Build main A section (7 bars), then create two 1-bar endings:
form.create_section "A" measures=8 # full 8-bar snapshot
# Build "A_end1" voices (1-bar turnaround back to A):
form.create_section "A_end1" measures=1
# Build "A_end2" voices (1-bar final cadence):
form.create_section "A_end2" measures=1
# Attach endings — pass 1 uses A_end1, pass 2 uses A_end2:
form.set_ending section="A" pass=1 ending_section="A_end1"
form.set_ending section="A" pass=2 ending_section="A_end2"
form.repeat_section "A"
form.build # LilyPond: \repeat volta 2 + \alternative; MIDI: correct ending per pass
Multi-section form with different voices per section (AABB, AABA, etc.):
# Each section gets its own voices. The pattern: build → snapshot → delete → rebuild → snapshot.
# harmony.set_bars must be reset per section; it is section-local.
# --- A section ---
harmony.set_key "D major"
harmony.set_bars "1:D 2:G 3:D 4:A 5:D 6:G 7:A 8:D"
voice.create "melody" "..." # 8 bars
harmony.comp target_voice="chords" style="on_beat" octave=3
voice.create "bodhran_bar" "G2/q G2/q G2/q G2/q"
voice.repeat "bodhran_bar" times=8 target_voice="bodhran"
voice.delete "bodhran_bar"
score.assign_instrument "melody" "flute"
score.assign_instrument "chords" "accordion"
score.assign_instrument "bodhran" "drums"
form.create_section "A" measures=8
form.repeat_section "A" # AABB: queue the second A now
# --- Transition: delete only the voices that change; keep shared ones (e.g. bodhran) ---
voice.delete "melody"
voice.delete "chords"
# --- B section ---
harmony.set_bars "1:G 2:D 3:G 4:A 5:G 6:Em 7:A 8:D"
voice.create "melody" "..." # 8 bars of new material
harmony.comp target_voice="chords" style="on_beat" octave=3
score.assign_instrument "melody" "flute"
score.assign_instrument "chords" "accordion"
# bodhran is still in scope — captured in B section too
form.create_section "B" measures=8
form.repeat_section "B" # AABB: queue the second B
form.build # assembles A(8) + A(8) + B(8) + B(8) = 32 bars
# All voices (melody, chords, bodhran) replaced with full 32-bar assembled versions
Key rules: delete only voices that change between sections; shared voices (drums, bass ostinato) stay in scope and are captured in every section automatically. harmony.set_bars is reset per section — form.build offsets and merges them into the final score.
Ritardando on final bars:
# Assume 24-bar piece at 120 BPM — slow to 60 over the last 3 bars:
score.set_tempo_change start_bar=22 end_bar=24 to_bpm=60 curve="exponential"
# exponential feels more natural than linear for a rit.
Canon at the octave:
voice.create "subject" "C4/q D4/q E4/q G4/q E4/q F4/q G4/q C5/q"
voice.pad_to_measure "subject" start_measure=3 target_voice="answer"
transform.transpose "answer" "P8" "down" target_voice="answer_low"
Motivic development (build a piece from a short cell):
# 1. Define a 4-note cell and save it as a motif
voice.create "cell" "C4/q E4/q G4/q B4/q"
motif.save "cell" startNote=0 endNote=4
# 2. Derive transformed variants — each becomes its own voice
voice.from_motif "cell_inv" motif="cell" transform="invert" axis="C4"
voice.from_motif "cell_retro" motif="cell" transform="retrograde"
voice.from_motif "cell_up5" motif="cell" transform="transpose" interval="P5"
# 3. Augment for a slower countermelody (double all durations)
transform.augment "cell" factor="2/1" target_voice="cell_slow"
# 4. Build voices by repeating cells — stagger entry for canon/imitation effect
voice.repeat "cell" times=4 target_voice="subject"
voice.repeat "cell_inv" times=4 target_voice="counter"
voice.pad_to_measure "counter" start_measure=2 # answer enters 1 bar later
# 5. B section uses retrograde or transposed variant as new material
voice.repeat "cell_retro" times=4 target_voice="b_melody"
voice.repeat "cell_up5" times=4 target_voice="b_counter"
# 6. Assemble into form
form.create_section "A" measures=4
# (swap in B voices here)
form.create_section "B" measures=4
form.repeat_section "A"
form.build # A(4) + B(4) + A(4) = 12 bars, all derived from the original cell
# Clean up ALL scratch voices before form.create_section or export
# Any voice still in scope will become a staff in the PDF
voice.delete "cell"
voice.delete "cell_inv"
voice.delete "cell_retro"
voice.delete "cell_up5"
voice.delete "cell_slow"
The key idea: one cell generates all the material. The transforms (invert, retrograde, transpose, augment) are the compositional engine — not just decoration.
For genre-specific scaffolding (Jazz, Bossa Nova, House, Reggae, Blues, Hip-Hop, Afrohouse, Traditional Irish), read skills/music-composition/genre-reference.md.
e eighths — the renderer adds the lilt.harmony.set_bars chords if set, otherwise falls back to harmony.chord_progression. Always set the key before calling walking_bass (needed for approach notes).harmony.set_bars has been called, chord symbols are automatically rendered above the staves as a \new ChordNames staff. No extra step needed; just export after setting bar chords.form.build — call harmony.set_bars separately for each section before calling form.create_section. form.build automatically merges and offsets each section's chord map to the correct absolute bar numbers in the assembled score. You do not need to manually re-number bars across sections.harmony.comp styles: quarter_stabs (beats 2&4 — jazz default), on_beat (all beats — rock/gospel), eighth_pump (8 per bar — funk), shell_voicings (root+7th on 2&4 — sparse jazz), charleston (dotted-quarter+eighth+quarter+rest — classic jazz). Use octave 3 for piano/Rhodes, octave 4 for guitar.~tie) are for notes that span bar lines or durations not expressible as a single value. The tied notes must have the same pitch; MIDI renders them as a single sustained note.harmony.set_bars chord symbols use the same quality syntax as ChordSymbol.parse: Cm7, F7, Ebmaj7, Am7b5, Ddim7. Bare root (e.g. Eb) means major.target_voice on all transforms is optional — omit to get {name}_transposed etc.transform.transpose transposes all pitches in the chord; transform.augment scales the chord's duration. transform.invert does not invert chord pitches — chords pass through unchanged.score.assign_instrument "kick" "drums" forces channel 9 automatically.harmony.set_key must be called before harmony.harmonize, harmony.diatonic_transpose.form.build rest-fills missing voices automatically — if "bass" exists in A but not B, B gets whole-note rests.form.build, voices are replaced in-place — no _form suffix, no stale originals.rules.check is advisory — violations are warnings/suggestions, never blockers. Percussion voices are skipped.harmony.set_key before exporting for correct notation and DAW display.harmony.set_key is set), and section markers (when form.build is used — DAW timeline shows "intro", "verse", "chorus" etc. at the correct bar positions).cc:expr:N events at phrase peaks and valleys rather than only at the start of a voice.voice.add_expression_curve is bar-absolute — bar numbers are relative to the final assembled score, not the section. In a form.build workflow, add curves after form.build using the assembled bar positions (e.g., if section B starts at bar 9, set start_bar=9). Curves added before form.build keep their original bar numbers, which is correct for section A but wrong for a repeated section B.from_value=20 to_value=90 curve=exponential), strings at bar 10, cello at bar 12. The exponential shape (quadratic ease-in: slow start, fast finish) suits crescendos; combine with a second curve going back down for a complete swell arc.