@autoaiai/aicut 0.1.11 → 0.1.13

package/skills/anidoodle/references/music/sounds-licenses.md

@@ -1,69 +0,0 @@
-# Sound pack licenses
-
-The sound pack is a separate download, installed by `engine/tools/soundfetch.mjs`. Nothing in this
-repository is licensed here; this page is what you need to carry if you redistribute the pack or a
-film made with it. The short version also lives in the repository's `NOTICE`.
-
-Two sources, two licenses. The instruments need nothing. The rooms need a line of credit.
-
-## Instruments: CC0
-
-All recordings from the Versilian Community Sample Library (VCSL).
-
-- Source: https://github.com/sgossner/VCSL
-- License: CC0 1.0 Universal (public domain)
-
-CC0 asks for nothing: no attribution, no licence notice, no share-alike, no restriction. Attribution
-is still welcome and the anidoodle page names the library.
-
-Eight instruments are taken from it:
-
-| Instrument id | Recording | VCSL folder |
-|---|---|---|
-| `piano` | Grand Piano, Steinway B | `Chordophones/Zithers/Grand Piano, Steinway B` |
-| `piano.upright` | Upright Piano, Knight | `Chordophones/Zithers/Upright Piano, Knight` |
-| `harp` | Concert Harp | `Chordophones/Composite Chordophones/Concert Harp` |
-| `marimba` | Marimba | `Idiophones/Struck Idiophones/Marimba` |
-| `vibes` | Vibraphone | `Idiophones/Struck Idiophones/Vibraphone` |
-| `glockenspiel` | Glockenspiel | `Idiophones/Struck Idiophones/Glockenspiel` |
-| `bell.tubular` | Tubular Bells 1 | `Idiophones/Struck Idiophones/Tubular Bells 1` |
-| `timpani` | Timpani | `Membranophones/Struck Membranophones/Timpani 2` |
-
-The exact WAV taken from each folder, and its sha256, are in the pack's own `LICENSES.md` next to
-`manifest.json`. That file is the authority; this table is the map.
-
-## Rooms: CC BY 4.0
-
-Recorded impulse responses from OpenAIR, the Open Acoustic Impulse Response library, Audio Lab,
-University of York: https://www.openair.hosted.york.ac.uk/
-
-- License: Creative Commons Attribution 4.0 International (CC BY 4.0)
-- The B-format (W X Y Z) recordings were decoded to stereo as L = W + Y, R = W - Y
-
-CC BY 4.0 requires attribution. Anyone redistributing the pack, or a film rendered with one of these
-rooms, must carry this credit:
-
-> Recorded impulse responses from OpenAIR, the Open Acoustic Impulse Response library, Audio Lab,
-> University of York (https://www.openair.hosted.york.ac.uk/), licensed CC BY 4.0.
-
-Credit each room where you can:
-
-| Room id | Room | URL |
-|---|---|---|
-| `small-room` | Arthur Sykes Rymer Auditorium, University of York | https://www.openair.hosted.york.ac.uk/?page_id=425 |
-| `recital-hall` | Jack Lyons Concert Hall, University of York | https://www.openair.hosted.york.ac.uk/?page_id=571 |
-| `live-room` | Genesis 6 Studio live room, University of York | https://www.openair.hosted.york.ac.uk/?page_id=483 |
-
-The other three things CC BY 4.0 allows are worth knowing: commercial use, modification (so
-processing the impulse response does not break it), and distribution. The one condition is the
-credit above.
-
-## Where this comes from
-
-`anidoodle-research/sound-pack/CONTRACT.md` sets the rules for version 1: FLAC, 48 kHz, 24-bit, the
-pack under 200 MB, and CC0 or public domain only. The three OpenAIR rooms are the exception the
-amendment of 2026-10-04 (b) authorized once the microphone was heard to sell the recordings. They
-are the only CC BY material the skill ships, and they are the only part that needs a credit line.
-
-Reproduce the pack and the licences with `node tools/soundpack.mjs`, which writes the per-file
-`LICENSES.md` from the same manifest the engine reads.

package/skills/anidoodle/references/music/styles/lofi-electronic.md

@@ -1,112 +0,0 @@
-# Style: lo-fi electronic (chill, upbeat, clean)
-
-> A vocabulary, not a song. The data is `VOCAB.lofiElectronic` in `engine/src/canvas-core/music/vocab.ts`
-> (print it: `node tools/music.mjs vocab lofiElectronic`). You compose the notes with
-> [../compose.md](../compose.md). Our launch film was scored in this style (`launchLofi3`). That
-> score is ours and is in the novelty corpus. Never reuse its progression, hook, rhythms, drums or
-> form.
-
-## The sound
-
-Chill but **upbeat**, **clear**, **not noisy**:
-- a warm pad that breathes with the kick;
-- a plucked lead through a dotted-eighth ping-pong delay;
-- a high counter-line;
-- a round sine sub;
-- a bouncy, clean kit.
-
-**Never:**
-- synth electric piano;
-- vinyl crackle;
-- bit-crush;
-- a dark master low-pass;
-- heavy tape.
-
-Those read as dusty lo-fi hip-hop (the `lofi` style), not clean electronic.
-
-| Slot | Voice | How it's made | Stem RMS target |
-|---|---|---|---|
-| chords | `warmPad` | 7 PolyBLEP saws, spread 0.55 x +-14 cents, slow per-voice drift, 12 dB low-pass at 2.6 kHz, 0.5 s attack, 1.4 s release | -22 dB |
-| lead | `softPluck` | triangle + square an octave down, filter snaps open, 0.42 s decay, dotted-eighth ping-pong (feedback 0.28, mix 0.24) | -15.5 dB |
-| counter | `softPluck` | brighter, panned 0.35, more delay and room | -20 dB |
-| bass | `sub` | sine + a whisper of the 2nd harmonic, low-passed at 150 Hz | -18.5 dB |
-| kick / snare / ghost / hat | kit | a bouncy kick, a snare, rim ghosts about 11 dB under, 16th hats | -14.5 / -19 / - / -24.5 dB |
-| arp, perc | `softPluck`, `hat` | optional extra layers | - |
-
-Alternates:
-- lead: `bell`, `mallet`, `guitar`;
-- chords: `strings`, `piano`;
-- bass: `warm`.
-
-Bus:
-- the kick ducks the chords and the counter (depth 0.38, release 0.24 s);
-- tape is barely there (2.5 cents of wow);
-- clean master at **-14 LUFS**, true peak **<= -1 dBTP**.
-
-The stem targets were calibrated on a listened score, and they are the only calibrated targets.
-
-## Grooves
-
-`bounce`, `halfTime`, `broken`, `skip`, `pulse`, `build`. Pick one per section role (`main`,
-`half`, `build`) and set `density` and `variation`. The seed varies every bar. Swing 0.52-0.56.
-
-## Harmony language (write your own progression)
-
-- **Modes:** major, dorian, lydian, aeolian, mixolydian.
-- **Qualities:** 7ths, 9ths, 11ths, 13ths, add9, sus2 and sus4. Keep bare triads for the landing.
-- **Tendencies:**
-  - loops of 2, 4 or 8 bars;
-  - ii-V motion;
-  - drifting IV-iii-vi motion;
-  - modal vamps (i-IV in dorian, I-II in lydian);
-  - a borrowed iv or bVII for a wistful turn.
-- **Harmonic rhythm:** one chord per bar or two. A change may anticipate the bar by an eighth.
-- **Voicing:** rootless 3-4 note voicings between about E3 and E4, moving by step. The sub owns the
-  root.
-- **Tension:** sus4 and 13ths lean, maj7#11 floats, 7b9 or a borrowed iv aches. Release by stepping
-  the leaning voice down.
-
-## Melody rules (write your own motif)
-
-- Lead in C4-A5, counter an octave above.
-- Short arches, with the high point late in the phrase.
-- 1-3 notes per beat, with rests so the echo can answer.
-- A 2-5 note cell with its own rhythm: repeat it, then vary it (a step up, the last interval
-  inverted, displaced by an eighth).
-- Call and response between the lead and the counter.
-
-## Arrangement grammar
-
-A typical shape: intro -> groove -> hook -> (half-time or breakdown) -> drop on home -> outro. Add
-or remove one layer at a time. Put the drop (everything back, home chord) on the picture's strongest
-beat.
-
-**Transitions:**
-- a fill bar;
-- a half-time bar;
-- drums out for a breakdown.
-
-**Endings:**
-- the home chord held, with the motif's last note;
-- a clean button.
-
-## Mixing: balance by stem RMS, then master
-
-```
-node tools/music.mjs check <file>.ts#<export>              # the whole gate, stems included
-node tools/music.mjs stems <file>.ts#<export>              # the stem table alone, while you balance
-node tools/music.mjs render <file>.ts#<export> out.mp3 --stems
-```
-
-A stem is a part's own signal:
-- unmastered, pre-room, after its gain and the duck;
-- measured on its **mid** ((L + R) / 2, after its pan, 30 Hz high-passed) as the RMS of the 20 ms blocks where it
-  plays: blocks under -70 dBFS never count, then blocks more than 20 dB under the part's own level drop out (gated
-  like loudness), at 24 kHz. (The v1 meter read the left channel sample by sample above -80 dB; every target was
-  converted at the sound-v2 merge on the render it was set on, so the balance did not move.)
-
-Fix a flagged part with `levels`, never with the master.
-
-**The lesson:** integrated LUFS only says how loud the whole mix is. On the launch score the mix read
-a healthy -14 LUFS while the sub sat **7-10 dB too hot**. `tools/music-unit.mjs` keeps this as a
-regression: a sub 9 dB hot still masters to -14.0 LUFS, and only the stem meter flags it.

package/skills/anidoodle/references/music/styles/music-box.md

@@ -1,57 +0,0 @@
-# Style: music box
-
-> The original anidoodle music recipe, kept below for its SOUND numbers (the `musicBox` and `bell`
-> instruments implement them) and its scar story (why the guards exist). It is one style among the
-> vocabularies in `engine/src/canvas-core/music/vocab.ts` (`VOCAB.musicBox`), not a default: there is
-> no default style. Its fixed composing choices (one progression, one scale, one accompaniment
-> figure) were rewritten below as ranges: the composer writes the notes ([../compose.md](../compose.md)).
-> Its ban list is the rule for THIS style only; for other styles the bans became measured guards.
-
----
-
-# The music recipe
-
-> Doctrine earned making MECHANICAL LEPIDOPTERA, the worked example in `example/`. Every rule
-> below was paid for on that film; where a mistake was the director's own it says so, because
-> a rule with its scar attached is one people keep.
-
-**The mistake.** Music made in code is the hardest thing in this skill, because the one making it cannot hear it. My spec asked for a music box that "asks a question and ends unresolved on the fifth", a tick "thinning into a felt flutter", and "a bowed pad from the take-off". Every one of those is a good sentence and together they are the recipe for what the client called a ghost crying: slow attacks, sustained tones, a phrase left hanging for eighteen seconds, reverb. It was nobody's bug. It was an adjective brief, and adjective briefs produce atmosphere, and synthesized atmosphere is a haunted house.
-
-**The rule: never "make it good". Ship this.**
-
-**Banned, all of it, always:** minor keys and modes; any sustained pad, bowed or blown tone, drone or sub-bass; attack times over 10 ms; reverb tails, feedback delays, convolution; detuning, pitch glides, vibrato; filtered noise "wind" or swells; a phrase left unresolved for more than four bars; a gap with nothing sounding for more than one beat once the music has started.
-
-**The instrument: a music box / celesta pluck.** Per note at frequency `f`:
-- partials: sine at `f` gain 1.0, `2f` 0.35, `3f` 0.12, plus a strike "ping" at about `5.4f` gain 0.08 that dies in 60 ms;
-- envelope: 2 ms linear attack, then pure exponential decay, no sustain. Time constant `tau = 0.45 * sqrt(440 / f)` seconds, so high notes are short and sparkly and low notes ring a little;
-- velocity: strong beats 1.0, others 0.7, grace notes 0.45.
-- bass: same pluck one to two octaves down with only the first two partials, gain 0.5. It is still a pluck. It never sustains.
-- "room": ONE early reflection, 30 ms late, 14 dB down. No tail.
-
-**Key, register, harmony.** Any major, lydian, dorian or minor key you choose for the mood (a minor music box is a real, uncanny sound). Melody between C5 and C7, bass C3 to G3. Harmonic rhythm one chord per bar; the progression is yours (plagal and deceptive turns read well; the original film used the plainest possible one and said "nothing cleverer", which is a choice, not a rule). For a lift at the payoff, move the key up a step.
-
-**Melody.** Chord tones plus the mode's scale, mostly steps, leaps onto chord tones. Four-bar phrases in pairs: the QUESTION ends off the tonic, the ANSWER lands on it, **and the answer follows immediately.** Your own motif, with its own rhythm.
-
-**Rhythm.** At 120 bpm and 30 fps a beat is 15 frames and a straight eighth is 7.5 frames, which does not exist. Use triplet eighths, 5 frames: 12/8, which is what a music box plays anyway. Accompaniment is an arpeggio figure you design on the triplets (choose its order of chord tones and its cycle length; a 4-note cell over 3-note groups makes a gentle hemiola); melody mostly on beats. Every onset is a row in a cue table with its FRAME, and the module asserts at load that each sample index equals `round(frame / fps * sampleRate)`.
-
-**Sound events ride on top, quietly:** a tick is a 2 ms noise burst through a band-pass at 3 kHz, the tock at 2 kHz, 20 dB under the melody. A landing is one soft pluck on its beat. The DING is a bell on the tonic (partials 1, 2, 3, 4.2, tau 1.2 s, doubled an octave down) on the final downbeat, with at least four beats of film after it.
-
-**Arc without pads.** Act by act: melody alone; add bass; add the arpeggio; add octave sparkle; bell. More notes, never longer notes.
-
-**Master.** Normalise to a peak of -3 dBFS, soft-clip with `tanh`, RMS near -18 dBFS.
-
-**Verification, since you are deaf.** Print the score as text, bar by bar, note names and chords, so a human can read it. Measure from the DECODED file: peak, RMS per bar, longest silence, and spectral centroid (a bright pluck score sits well above 1.5 kHz; a number near 400 Hz means a pad got in). Then the only test that counts: **one 8-second sample to the client before you score the film.** It is the music's look-still. We skipped it and scored sixty seconds of the wrong thing.
-
----
-
-## What is verified here, and what is not
-
-Flagged by the director when this doctrine was written, and carried forward unchanged because it
-still applies: the "what went wrong" above rests on the **client's description** of the render plus the
-director's own spec, which had asked for exactly the ingredients she named. The director could not
-hear it either. **The synthesis numbers are a good starting point, not a frozen answer** — confirm
-them with one human listening pass before you trust them on a new film.
-
-The related `doodle-film` skill's score advice (detune the loop flatter on each repeat, a hush and
-a heartbeat kick) is tuned for a wistful film and contradicts several bans above. Neither is wrong;
-they are for different moods. Say which mood you are in before you pick.