@hraness/dawg 0.3.0 → 0.4.1

package/CHANGELOG.md

@@ -4,6 +4,144 @@ All notable changes to dawg are recorded here. Versions follow [semantic version
 
 ## Unreleased
 
+## 0.4.1
+
+Sound previews now crossfade instead of clicking, reverb changes are heard in under 100 ms, and the rhythm editor and chord settings hold changes until you keep them. dawg.sh gains a security policy and security headers.
+
+### Previewing changes
+
+- **Edits crossfade instead of clicking.** When the loop (audition or song) swaps to a new render, the old and new buffers crossfade over 20 ms with equal-power gains at the same beat, so a change no longer steps the waveform. Playback only: renders and exports are byte-identical.
+- **`/euclid` and the chord settings stage while the loop plays.** `Space` in the rhythm editor or the menu's Chords section (`/menu chords`) starts the audition loop; changes are then staged and heard (`●`, `B staged N`, `E(5,16) ← E(4,16)`, `Db ← C`), `a` flips A/B, `Enter` keeps them as one undo step and `Esc` reverts with nothing written. In the Chords section the loop plays the track's chord phrase under the current settings. With the loop off, both commit at once as before; their key footers and `?` panels use the shared audition keys.
+- **Reverb changes are heard faster.** The stem cache keeps each track's reverb input and tail apart from its dry signal, so a reverb mix (or mix lane) change re-mixes the cached tail instead of re-rendering the voice and the room. Key to audio scheduled on a 4-track song: reverb mix on a wavetable pad in context 140 → 64 ms median (p90 151 → 68 ms); a filter change in context 100 → 74 ms. The bus mix loops also moved out of the large render function, which the JIT optimized late. Renders stay byte-identical (a new test checks cached re-mixes against cold renders and recorded digests).
+
+### Security
+
+- **A security policy.** `SECURITY.md` and `https://dawg.sh/.well-known/security.txt` say how to report a vulnerability privately.
+- **dawg.sh sends security headers** on every route: a Content Security Policy limited to the site's own origin and its analytics hosts, HSTS, `nosniff`, `DENY` framing, a strict referrer policy and a permissions policy that turns off camera, microphone and location.
+- Dependabot also watches the site's dependencies (minor and patch updates, grouped).
+
+## 0.4.0
+
+dawg now has a full sound engine with Strudel's synth, effect and sample parameters, Strudel sample packs, wavetables (including ones made from your own audio), Euclidean drum rows and a pattern library, Orchid-style chords, sound previews with A/B, and one consistent set of keys across every screen. Projects from 0.3.0 open and sound the same, and the `dawg` SDK is 1.13.0 (every step additive within v1).
+
+### Previewing changes
+
+- **Audition loop in the edit menu.** `Space` loops the focused track (solo, or the whole mix with `c`), using its own notes over a region of at most four bars, or a short phrase by role when it has none (chord, riff, groove, or one held wavetable note).
+- **Staged edits with A/B.** While the loop plays, menu changes are staged and heard within about 100 ms (only the changed track re-renders, off-thread). `a` flips between the committed and the staged sound, `Enter` keeps all staged changes as one revision and one undo step, `Esc` reverts. Changed rows show `staged ← committed`. Edits from another window are re-applied under the staged ones.
+- The audition controller (`src/tui/audition.ts`) and the preview score builder (`src/audio/preview.ts`) are reusable by any screen; a preview renders byte-identical to the same bars of a full render.
+- **Hover to hear lists.** With the loop on, moving through a list (wavetables, instruments, drum kits, patterns, choice lists in `/menu`, and the `/kit` and `/pattern` pickers) plays the highlighted item on the loop. Hovers replace each other, fast moves skip straight to the latest item, and a pack sound that needs fetching shows `fetching…` without blocking the cursor. Enter chooses, Esc leaves with nothing changed.
+- **The agent can hear before it commits.** `preview_sound` renders a track, or candidate `set_fx`/`set_synth`/`set_wavetable`/… calls applied to a copy, over the audition loop's bars, and returns RMS/peak dBFS, spectral centroid, a description and a comparison with the current sound. It plays once in a quiet window; `/try agent off` keeps it silent.
+- **Audition visuals.** The menu title shows a level meter with a clip mark while the loop plays, and the focused cutoff, envelope or wavetable position row draws a one-line sketch of its response.
+- **`/try <sound command>`** hears a prompt command on the loop first: `/try fx reverb mix 0.6`, then `a` for A/B, Enter to keep (one undo step) or Esc to drop it.
+
+### Effects
+
+- **A full effects chain on every track**, in a fixed order: filter → dj filter → auto filter → vowel → bitcrush → distortion → tremolo → compressor → pan → phaser → chorus → leslie → post gain → delay → reverb. Each effect turns on with a good default and has a few simple parameters, with the rest under **advanced**.
+- **Filter** gains `type` lpf/hpf/bpf and a 12 dB, 24 dB or ladder slope. **Auto filter** sweeps the cutoff with a tempo-synced or free LFO (sine, triangle, square, saw, ramp, sample-and-hold) and can follow the input's envelope.
+- **Delay** turns on as a dotted-eighth stereo ping-pong with gentle feedback and a high-cut on the repeats. **Reverb** adds decay time, damping and pre-delay.
+- **Distortion** (ten curves, tone, mix, automatic gain compensation), **compressor** (threshold, ratio, knee, attack, release, make-up), **chorus**, **tremolo**, plus Strudel's dj filter, vowel, bitcrush/coarse, phaser, leslie and post gain.
+- Parameter names follow Strudel's (`lpq`, `delayfeedback`, `room`, `tremolodepth`…), and every numeric parameter can be automated (`automate distort-drive points 0:1 8:6`).
+- **`fx` prompt command**: `fx delay mix 0.3`, `fx filter type hpf cutoff 300`, `fx chorus preset wide`, `fx tremolo off`. The Effects menu lists the core effects first, with presets. The agent's new `set_fx` tool sets the same parameters.
+- **ZzFX sounds**: `instrument z_square` (and `z_sine`, `z_triangle`, `z_sawtooth`, `z_tan`, `z_noise`) with Strudel's ZzFX controls as synth parameters: `synth slide -4 pitchJump 300 pitchJumpTime 0.1 lfo 0.25 zcrush 0.3`, plus `zrand`, `curve`, `deltaSlide`, `zmod`, `zdelay`, `tremolo` and the `zap` preset. Clean-room, deterministic. SDK 1.10.0 (additive).
+- **Orbits and sidechain ducking**: `fx orbit 2` puts a track on orbit 2; `fx duck preset pump` on a kick dips every track on that orbit at each hit and lets it swell back (Strudel's `orbit`, `duckorbit`, `duckdepth`, `duckattack`, `duckonset`). SDK 1.9.0 (additive).
+- **Convolution reverb** (Strudel `ir`/`iresponse`): `fx reverb ir hall` (also `room`, `plate`, a pack sound or a project WAV), with FFT convolution; `fx ir off` returns to the algorithmic tail. SDK 1.12.0 (additive).
+- **Shared orbit buses**: `fx orbit shared on` sends a track to one delay and one reverb shared by its orbit, as Strudel orbits do.
+- **Raw ZzFX arrays**: `synth zzfx ,,129,.01,,.15,2`, SDK `zzfx([...])` and `set_synth { zzfx }` map ZzFX's positional array to the named controls.
+- **Strudel sample controls** on sampler voices: `loopBegin`/`loopEnd`, `clip` (`legato`), `fit`, `unit` "c"/"s", `loopAt`, `accelerate` and `squiz`, beside `begin`/`end`/`speed`/`loop`/`cut`. `/sample set brk fit on clip 1`, the per-voice menu and the agent's `set_sample` tool edit them. SDK 1.13.0 (additive).
+- Older projects and sessions load and sound exactly as before. The `dawg` SDK is 1.5.0 (additive).
+
+### Synth
+
+- **A full synth voice with Strudel's parameters.** `synth` on a track takes Strudel's names and aliases: sine, sawtooth, square, triangle, supersaw, pulse, additive `user` and white/pink/brown/crackle noise; `noise`/`density`; `unison`/`detune`/`spread`; pulse width with `pwrate`/`pwsweep`; eight FM operators (`fm`, `fmh`, an ADSR, `fmenv`, `fmwave`); ADSR; a pitch envelope (`penv`…); vibrato; and low-, high- and band-pass filters with envelopes and `ftype`.
+- **Presets**: pad, lead, pluck, bass, sub, acid, keys, bell, organ, strings, brass, wind, chip.
+- **`synth` prompt command**: `synth preset pad`, `synth lpf 800 lpenv 3`, `synth fm 4 fmh 1.5`, `synth reset`. The menu's Parameters section shows the preset and simple parameters, with every parameter under **advanced**. The agent's new `set_synth` tool sets the same parameters. Every numeric parameter has a `synth-<param>` automation lane.
+- Older projects and sessions load and sound exactly as before. The `dawg` SDK is 1.6.0 (additive).
+
+### Rhythm (Euclidean rows)
+
+- **Drum parts as generators.** A kit or oneshot sampler track can carry `rhythm` rows: `euclid("kick", 4, 16)`, `euclid("hat", 7, 16, 2, { swing: 0.15 })`, `grid("snare", "....X.......x...")`. dawg expands them into ordinary notes, so you and the agent edit four numbers instead of sixteen hits, and rendering, sync and diffs are unchanged. Patterns and rotation match Strudel's `euclid`/`euclidRot`.
+- **Torso T-1 parameters**: steps, pulses, rotate, division, repeats with time/pace/ramp (rolls that speed up, slow down, build or fade), velocity, accent, gate/legato, seeded probability, swing, nudge and per-pass cycles. The same seed always gives the same hits.
+- **`/euclid` editor** (also Rhythm in `/menu`): one row per voice with its step grid. Arrows nudge the selected parameter, Tab moves between parameters, digits type a value, Space auditions. Every change is one undo step.
+- **Prompt grammar**: `euclid kick 4 16`, `euclid hat 7 16 rotate 2`, `euclid hat swing 0.2`, `euclid snare off|freeze`, `grid snare ....X.......x...`.
+- The agent's new `set_rhythm` tool takes the same rows, and the agent prefers it to hand-placed drum hits.
+- Editing a generated lane by hand freezes that row into plain notes. Changing the loop length regenerates rows.
+
+### Chords
+
+- **Chord engine (`core/chords.ts`)** modelled on the Telepathic Instruments Orchid: four chord types (dim, min, maj, sus) with combinable 6, m7, M7 and 9 extensions, Key-mode diatonic chords for major, minor and the church modes, a voicing dial, voice leading to the nearest inversion, bass, and block, strum, arpeggio and harp performance.
+- **Progressions.** Eleven presets (I–V–vi–IV, ii–V–I, i–VI–III–VII, the Andalusian cadence, …) and pop, jazz, modal and classical styles that walk a weighted functional-harmony graph with a seed, so the same request always gives the same chords.
+- **Secret chords from the Orchid manual.** Two latched chord types now play the official table: dim+sus C5, maj+sus C+, min+sus Cm(add4), min+dim Cm(b6), maj+dim C(b6), maj+min C7♯9 (they replace dawg's earlier guesses).
+- **Slop perform mode**, Orchid's humanised timing: a block chord whose voices land a seeded fraction of up to 1/16 beat late (SDK 1.5.0, `perform: "slop"`).
+- **Pattern perform mode** with thirteen chord rhythms (eighths, offbeat, pop, charleston, bossa, skank, tresillo, oom-pah, …) by name or number: `/chords pattern 3`, `9` in play mode, the menu, `write_chords {perform: "pattern", pattern}` and SDK 1.7.0 `perform: "pattern", pattern: "bossa"`.
+- **Orchid bass modes.** Bass is now `off`, `chords`, `unison`, `single` or `solo`, as in the Orchid's bass menu: `B` in play mode cycles them, `/chords bass unison`, the menu, `write_chords {bassMode}` and SDK 1.7.0 `bass: "solo"`.
+- **Agent tools.** `suggest_progression` returns voice-led chords with names, numerals and bass; `write_chords` writes them to a track (and its bass to another) as block chords, strums, arpeggios or harp sweeps.
+- **SDK 1.3.0.** `chord("Cm7")` and `progression(["ii7", "V7", "Imaj7"], { key, perform })` expand to notes in `track.ts` files.
+- **Chord mode in play mode**, on by default (`auto`) for chord-capable tracks: every note key plays the song key's diatonic chord, voice-led from the last one. `1`–`4` latch dim/min/maj/sus, `5`–`8` latch 6/m7/M7/9, `0` clears, `-`/`=` turn the voicing dial, `9` cycles the perform mode, `B` toggles bass, `N` plays the suggested next chord and `Q` switches auto ⇄ manual. The header shows the mode, key, current chord and the suggestion. Recorded chords keep their strum or arpeggio and stay one undo step per bar.
+- **`/chords`** and a Chords section in `/menu` edit the same settings; **`key <tonic> <mode>`** sets the song key.
+
+### Wavetable synth
+
+- **`instrument: "wavetable"`**: a band-limited wavetable oscillator (per-octave mipmaps, smooth frame interpolation) with Strudel's parameter names: position `wt`, a position envelope (`wtenv`, `wtattack`, `wtdecay`, `wtsustain`, `wtrelease`), a position LFO (`wtrate`, `wtdepth`), `warp`/`warpmode` (`asym`, `bendp`, `bendm`, `bendmp`, `sync`, `quant`) and `wtphaserand`. `wt` is automatable.
+- **Tables**: four built-ins that work offline (`basic`, `pwm`, `formant`, `harmonics`) and Strudel's `wt_` sounds (`wt_digital:2`) from the new `uzu-wavetables` pack, fetched once and pinned by sha256 like any pack sound.
+- **`/wt <table|0..1|list>`**, one command per parameter (`/wtenv 0.5`, `/warpmode bendp`), the table picker and parameters in the menu's Parameters section, and the agent's `set_wavetable` tool. Play mode plays wavetable tracks.
+- It runs through the new synth voice, so ADSR, filter envelopes, FM, unison/detune and vibrato apply to wavetable tracks too.
+- SDK 1.8.0 (additive): `wavetable("basic", { wt: 0.4 })` as a track's `instrument`, and `automation: { wt: [...] }`.
+
+### Wavetables from audio
+
+- **`make_wavetable`**, a new agent media tool (`dawg media wavetable <file> <name>` from the shell): turns any audio in the project (a download, a stem, an imported sample) into a 2048-sample-frame wavetable at `tracks/<slug>/wavetables/<name>.wav`. Pitched material is sliced into single cycles at the detected pitch; vocals, pads and noise become spectral snapshots. It picks the most stable tonal region by default, phase-aligns and normalises the frames, and tells the agent how the timbre moves across the table (for example "brightens, very smooth morph") so it can choose a position or envelope. Output is deterministic.
+- **Project tables play like any other table**: `/wt vox.wav`, the menu's table picker (project tables are listed first), `set_wavetable` with the path, and `wavetable("./wavetables/vox.wav")` in `track.ts`. They are pinned by sha256; a changed file plays with a warning, a missing one is reported with a fix.
+- SDK 1.11.0 (additive): `wavetable()` accepts a project `.wav` path.
+
+### Sample packs
+
+- **Strudel's sample ecosystem, read through its public manifest format** (clean-room, no Strudel code). Built-in catalog: tidal-drum-machines, dirt-samples, uzu-drumkit, VCSL, piano, mridangam, emu-sp12 and two General MIDI soundfonts, each with its license.
+- Files download lazily on first use over HTTPS and are cached offline. A sampler voice can point at `pack:<pack>/<sound>[:<n>]`, and first use pins its sha256, URL and license. SDK 1.1.0 (additive).
+- `/pack list|add|info|remove|use`, `/kit <bank>`, a sounds list in the menu, and agent tools `list_packs`, `search_sounds` and `use_sound`.
+- Renders credit the packs they use, and projects using CC-BY or CC-BY-SA packs get a `CREDITS.md`.
+
+### Sample packs: bank nicknames and cache sizes
+
+- **Bank nicknames.** Strudel's drum-machine nicknames (`TR909`, `tr808`, `Linn`, `DMX`, `SP12`, `MPC60`, … from its `tidal-drum-machines-alias.json`) work in `/kit`, `/pack use`, `pack:` refs in `track.ts`, the agent's `use_sound` and a new **Strudel banks** list in the menu's drum kits. dawg ships a snapshot and refreshes it with the manifest. `909`, `808`, `linn` and the other short names work as before, and pins keep the full bank name.
+- **`/pack cache`** shows disk used by pack downloads and decoded audio against their caps; `/pack cache prune [size]` and `/pack cache clear` evict least recently used files and keep the open project's sounds.
+- Pack downloads are now capped at 2 GiB (they were uncapped) and decoded audio at 1 GiB per project (was 512 MiB), overridable with `DAWG_PACKS_CACHE_MAX` and `DAWG_ASSETS_CACHE_MAX`. Eviction never removes a file the open project uses; an evicted pack file re-fetches by its pinned sha256.
+
+### Drum patterns and kits
+
+- **Pattern library.** 31 starting grooves (house, techno, boom bap, trap, drill, dnb, reggaeton, afrobeats, bossa nova, samba, garage, jersey club, footwork, funk and more), each a set of rhythm rows you keep editing as parameters. `/pattern` opens a picker that plays one bar of the pattern under the cursor; `/pattern <name> [keep-tempo|tempo]` applies one and moves the tempo into the pattern's range when needed. Also under Sounds → Drum patterns in `/menu`, and `pattern("boom-bap")` in `track.ts`.
+- **Synthesized kits.** `kit: "syn808" | "syn909" | "acoustic" | "lofi" | "electro" | "trap"` on a drum track, or `/kit <name>`, chooses an offline drum synth sound. Tracks without a kit sound exactly as before.
+- **One kit picker.** A bare `/kit` now opens a picker of every kit, synth kits first and then the sample kits from packs (it used to apply the 909 sample kit directly; `/kit 909` still does). `/kit syn909` on a sampler kit turns it back into a synth kit.
+- The agent's new `list_drum_patterns`, `apply_drum_pattern` and `set_drum_kit` tools, and it starts genre grooves from a pattern.
+- SDK 1.4.0.
+
+### Menus (reorganised)
+
+- **`/menu [section]` or Ctrl-K** opens six plain sections, most used first: **Sound**, **Effects**, **Rhythm**, **Chords**, **Mix & automation** and **Project**. The old section names (`parameters`, `sounds`, `track`, `automation`, `transport`) still open the matching place.
+- Rows show a plain label and the value with its unit (s, Hz, oct, st, dB, BPM, bars); a line under the list describes the focused row and shows, dimmed, the prompt command it runs. Every change is one receipt and one undo step; `x` resets a value to its default.
+- New prompt commands behind the menu: `automate <lane> points <b:v>...`, `automate <lane> remove <beat>`, `track name <text>` and `meter <n>`.
+
+### One key grammar
+
+- **Every picker, menu, editor and panel uses the same keys**: `↑↓`/`j k` move, `←→`/`h l`/`- +` adjust, Enter opens or confirms, Space auditions or toggles, `/` filters, Esc goes back one level (filter or typed value first), `?` shows the keys for the current screen. Digits type a value only where one is focused.
+- **A one-line key footer on every screen** that fits 80 columns; when narrower it drops the middle and keeps `esc` and `? keys`.
+- **`?` panel**: the keys for the screen underneath, drawn over it; on an empty prompt it lists the prompt keys and the three ways in (type a request, Ctrl-P, Ctrl-K). In play mode it also holds velocity, grid, click, count-in and the chord settings.
+- **Play mode header** keeps range, record state and the chord at a glance (`AUTO C major · Dm (ii) · next G`); chord mode adds a legend row for the number-row latches.
+- The `/euclid` editor is titled **rhythm** and its rows use the shared keys; pickers report the highlighted row on every move (`pick-move`), the hook live previews (and a future audition controller) listen on.
+
+### Help
+
+- **Sentences reach the agent**: a request that merely starts with a command word (`add a walking bass in A minor`, the prompt's own example, or `pan the hats left`) goes to the agent instead of answering with the command's usage. Short or numeric misuse (`pan 3`, `add H4 at 0`) still gets the usage hint locally.
+- **`/help` is a short task guide** (start here, play notes, make drums, shape the sound, chords) pointing to Ctrl-K, Ctrl-P and `?`; `/help all` is the full reference and `/help music|session|window|keys` shows one group. The reference now lists `fx`, `synth`, `/pattern`, `/kit` and `/pack`.
+- **Typos get the nearest command**: `/clik on` answers `unknown command /clik · did you mean /click? · /help` instead of a bare "unknown". Entering play mode says `? keys · esc leaves` instead of listing four keys.
+- **Bare-word slips are caught too**: `tempoo 90` answers `did you mean tempo 90?` without a model call when the rest parses as that command's arguments; sentences still go to the agent.
+- **First run names the three ways in**: an empty track reads `<track> · empty · type a request · ctrl-p play · ctrl-k menu` (narrow windows keep `add C4 at 0 to start`).
+- **The play `?` panel's chord line uses the header's words** (`next G`, `(assumed)`, `voicing +1`) instead of `→`, `?` and `inv`.
+
+### Fixed
+
+- Rhythm rows with options but no rotate printed as `euclid("hat", 8, 16, { … })`, which the SDK read as a rotate and rejected. The printer now writes rotate 0, and the SDK accepts options in that position.
+- Gateway web searches no longer count the search fee twice in the spend line and ledger. The gateway's reported cost already includes it. A real Exa search response is now a test fixture.
+
 ## 0.3.0
 
 dawg projects are now plain TypeScript files that you, an agent or another window can edit, with sampler tracks, local media tools, a computer-keyboard play mode and menus for every edit by hand.