livepilot 1.10.7 → 1.10.8

package/livepilot/skills/livepilot-sample-engine/references/sample-techniques.md

@@ -1,131 +0,0 @@
-# Sample Technique Catalog — 29 Recipes
-
-## Category 1: Rhythmic Sampling
-
-### slice_and_sequence (Surgeon)
-Classic MPC workflow: load loop, slice on transients, sequence with MIDI.
-**Material:** drum_loop, full_mix | **Steps:** load -> slice -> get slices -> create clip -> add notes
-
-### vocal_chop_rhythm (Alchemist)
-Chop vocal into syllable-length slices, trigger as staccato rhythm. Burial-inspired.
-**Material:** vocal | **Steps:** load -> slice by region -> MIDI staccato -> Auto Filter
-
-### micro_chop (Alchemist)
-1/32 slices, varied velocity, slight timing offsets. J Dilla micro-timing.
-**Material:** any loop | **Steps:** load -> manual slice -> dense 1/32 MIDI
-
-### stab_isolation (Surgeon)
-Isolate single chord stab, crop, retrigger rhythmically. DJ Premier style.
-**Material:** full_mix, instrument_loop | **Steps:** load -> classic mode -> crop -> MIDI
-
-### euclidean_slice_trigger (Alchemist)
-Map Simpler slices to Euclidean rhythm for polyrhythmic texture.
-**Material:** drum_loop, vocal, instrument_loop | **Steps:** load -> slice -> Euclidean pattern
-
-## Category 2: Textural Transformation
-
-### extreme_stretch (Alchemist)
-Paulstretch-style: Texture warp at 10-50x, reverb wash. Stars of the Lid territory.
-**Material:** any | **Steps:** load -> warp 64 beats -> Reverb 60-80% wet
-
-### drum_to_pad (Alchemist)
-Reverse + extreme stretch + reverb = drum hit becomes ambient pad.
-**Material:** drum_loop, one_shot | **Steps:** load -> reverse -> stretch -> Reverb
-
-### reverse_layer (Alchemist)
-Reversed sample as pre-echo swell or ghostly texture.
-**Material:** vocal, instrument_loop, one_shot | **Steps:** load -> reverse -> Delay
-
-### granular_scatter (Alchemist)
-Grain Delay as granular engine — scatter grains for cloud textures. Amon Tobin.
-**Material:** vocal, instrument_loop, texture | **Steps:** load -> Grain Delay -> Reverb
-
-### spectral_freeze (Alchemist)
-Crop to tiny region (50-200ms), extreme stretch = spectral freeze drone.
-**Material:** vocal, instrument_loop, full_mix | **Steps:** load -> crop -> stretch 64 beats
-
-### tail_harvest (Alchemist)
-Resample only the reverb/delay tail as independent texture.
-**Material:** any | **Steps:** load -> Reverb 100% wet -> level for subtle layer
-
-## Category 3: Melodic/Harmonic
-
-### key_matched_layer (Surgeon)
-Transpose to song key, EQ carve, blend behind existing elements.
-**Material:** instrument_loop | **Steps:** load -> transpose -> EQ -> volume -6 to -10dB
-
-### vocal_harmony_stack (Surgeon)
-Pitch-shifted vocal layers — Bon Iver Prismizer approach.
-**Material:** vocal | **Steps:** load -> duplicate -> transpose +3/+5/+7 -> blend
-
-### counterpoint_from_chops (Alchemist)
-Create countermelody from rearranged melodic fragments. Four Tet.
-**Material:** instrument_loop, vocal, full_mix | **Steps:** load -> slice by beat -> program melody
-
-### chord_stab_extraction (Surgeon)
-Isolate chord from full mix, crop, retrigger. Classic house/disco.
-**Material:** full_mix, instrument_loop | **Steps:** load -> slice -> crop -> rhythmic pattern
-
-## Category 4: Drum Enhancement
-
-### break_layering (Surgeon)
-Layer drum break under programmed drums. High-pass to avoid kick clash.
-**Material:** drum_loop | **Steps:** load -> warp 16 beats -> EQ HP 200-400Hz -> volume -10 to -15dB
-
-### ghost_note_texture (Alchemist)
-Heavy filter + low volume = barely audible ghost-note layer.
-**Material:** drum_loop | **Steps:** load -> Auto Filter BP 1-4kHz -> volume -18 to -24dB
-
-### transient_replacement (Surgeon)
-Layer one-shot transient over existing drums for punch.
-**Material:** one_shot | **Steps:** load -> classic mode -> MIDI on kick/snare hits
-
-### shuffle_extract (Alchemist)
-Extract groove timing from loop via slice positions, apply to MIDI.
-**Material:** drum_loop | **Steps:** load -> slice by transient -> read positions
-
-## Category 5: Vocal Processing
-
-### syllable_instrument (Alchemist)
-Each syllable = playable note. Vocal becomes a melodic instrument.
-**Material:** vocal | **Steps:** load -> slice by region -> program melody across slices
-
-### formant_shift_character (Alchemist)
-Shift formants for alien/robotic character. Transpose +/-12st.
-**Material:** vocal | **Steps:** load -> transpose -> Corpus for resonant body
-
-### vocal_freeze_drone (Alchemist)
-Sustain one vowel as ambient pad. Crop tiny region, extreme stretch.
-**Material:** vocal | **Steps:** load -> crop 100-300ms -> stretch 64 beats -> Chorus -> Reverb
-
-### phone_recording_texture (Alchemist)
-Burial signature: pitch down, lo-pass, ghost-level volume.
-**Material:** vocal, foley | **Steps:** load -> pitch -5 to -12 -> LP 800Hz -> volume -20 to -30dB
-
-## Category 6: Resampling Chains
-
-### serial_resample (Alchemist)
-Multi-pass destruction: Saturator + Grain Delay + Reverb, freeze, flatten, repeat.
-**Material:** any | **Steps:** load -> Saturator -> Grain Delay -> Reverb -> freeze -> flatten
-
-### parallel_resample (Alchemist)
-Duplicate, process one copy destructively, blend wet/dry.
-**Material:** any | **Steps:** load -> duplicate -> process duplicate -> blend -6 to -12dB
-
-### freeze_flatten_rechop (Alchemist)
-Freeze processed material, flatten to audio, re-slice the result. Recursive.
-**Material:** any | **Steps:** freeze -> flatten -> re-load into Simpler -> slice again
-
-## Category 7: Creative Constraints
-
-### one_sample_challenge (Alchemist)
-Build entire beat from one sample: kick, snare, hat, bass, pad all from slices.
-**Material:** any | **Steps:** load -> slice -> program full beat across slice pitches
-
-### found_sound_only (Alchemist)
-Non-musical field recordings as sole source material. Musique concrete.
-**Material:** foley | **Steps:** load -> slice -> EQ to isolate musical frequencies
-
-### reverse_engineering (Both)
-Recreate a reference track's texture by sampling and transforming similar elements.
-**Material:** full_mix | **Steps:** load -> slice by beat -> analyze structure

package/livepilot/skills/livepilot-sound-design-engine/SKILL.md

@@ -1,168 +0,0 @@
----
-name: livepilot-sound-design-engine
-description: This skill should be used when the user asks to "design a sound", "analyze a patch", "fix a static sound", "add modulation", "check my timbre", "improve a synth patch", or wants critic-driven sound design feedback and iterative patch refinement.
----
-
-# Sound Design Engine — Critic-Driven Patch Refinement
-
-The sound design engine analyzes synth patches, identifies timbral weaknesses, and iteratively refines them through a measured critic loop. Every change is evaluated against the before state.
-
-## The Sound Design Critic Loop
-
-### Step 1 — Build Patch Model
-
-Call `get_patch_model(track_index)` to build a PatchModel for the target track. The PatchModel maps every device on the track into typed blocks (oscillator, filter, envelope, lfo, saturation, spatial, effect) and classifies each as `controllable` or `opaque`.
-
-Read the response carefully:
-- `blocks`: ordered list of processing blocks with types and parameter names
-- `controllable_params`: parameters you can modify via `set_device_parameter`
-- `opaque_blocks`: third-party plugins where parameters may not map cleanly
-- `modulation_sources`: detected LFOs, envelopes, and macro mappings
-- `signal_flow`: how blocks connect (serial, parallel, or rack chains)
-
-See `references/patch-model.md` for the full PatchModel structure and native device block map.
-
-### Step 2 — Analyze
-
-Call `analyze_sound_design(track_index)` to run all sound design critics against the patch. The response contains an `issues` array with `critic`, `severity`, `block`, and `evidence`.
-
-Five critics run during analysis. See `references/sound-design-critics.md` for thresholds:
-
-- **static_timbre** — sound does not evolve over time, no movement
-- **no_modulation_sources** — no LFOs, envelopes, or automation detected
-- **modulation_flatness** — modulation exists but ranges are too narrow to hear
-- **missing_filter** — raw oscillator output with no spectral shaping
-- **spectral_imbalance** — too much energy in one frequency region, or gaps
-
-### Step 3 — Plan
-
-Pick the highest-severity issue. Call `plan_sound_design_move(track_index)` with the issue. The planner returns a single intervention:
-
-- `move_type`: one of the move vocabulary entries
-- `target_device`: device index on the track
-- `target_parameter`: parameter name or index
-- `target_value`: the new value
-- `rationale`: why this move addresses the issue
-
-Move vocabulary:
-- **modulation_injection** — add or increase LFO/envelope depth on a parameter
-- **filter_shaping** — adjust cutoff, resonance, or filter type
-- **parameter_automation** — create clip automation for time-varying timbral change
-- **oscillator_tuning** — adjust pitch, detune, waveform, or unison settings
-
-### Step 4 — Capture Before
-
-1. Call `get_device_parameters(track_index, device_index)` — save current parameter state
-2. Call `get_master_spectrum` — save spectral snapshot (if analyzer available)
-
-### Step 5 — Execute
-
-Apply the planned move using the appropriate tool:
-
-- `set_device_parameter` for direct parameter changes (filter cutoff, LFO rate, oscillator shape)
-- `toggle_device` for enabling/disabling processing blocks
-- `batch_set_parameters` when the move requires coordinated changes (e.g., LFO depth + rate together)
-- `set_clip_automation` for parameter automation moves
-- `find_and_load_device` when the fix requires adding a new device (e.g., adding an Auto Filter)
-
-Execute one move at a time. Verify before continuing.
-
-### Step 6 — Capture After
-
-Repeat the same measurements:
-
-1. Call `get_device_parameters(track_index, device_index)` — confirm the change took effect
-2. Call `get_master_spectrum` — save post-change spectral snapshot
-
-### Step 7 — Evaluate
-
-Call `evaluate_move(goal_vector, before_snapshot, after_snapshot)` where `goal_vector` is the compiled goal from Step 1 and snapshots contain `{spectrum: {...}, rms: float, peak: float}`. Read:
-
-- `keep_change` (bool): whether the change improved the sound
-- `score` (0.0-1.0): quality improvement magnitude
-- `timbral_delta`: what changed spectrally
-- `explanation`: human-readable summary
-
-### Step 8 — Keep or Undo
-
-If `keep_change` is `false`, call `undo()`. Explain what was tried and why it did not improve the sound.
-
-If `keep_change` is `true`, report the improvement. If score > 0.7, consider calling `memory_learn(name="...", type="device_chain", qualities={"summary": "..."}, payload={...})` to save the technique.
-
-### Step 9 — Repeat
-
-Return to Step 2. Re-analyze after each kept change. The critic list updates as issues are resolved. Continue until the user is satisfied or no high-severity issues remain.
-
-## Working with Opaque Plugins
-
-Third-party AU/VST plugins may report as `opaque` in the PatchModel:
-
-1. Check `get_plugin_parameters(track_index, device_index)` — some plugins expose parameters through the host
-2. If `parameter_count <= 1`, the plugin is dead or unresponsive. Call `delete_device` and suggest a native alternative
-3. If parameters are available but unnamed (Parameter 1, Parameter 2...), try `map_plugin_parameter` to identify them by ear
-4. Report opaque status to the user — sound design critics cannot fully analyze what they cannot inspect
-
-## Quick Sound Design Checks
-
-- **"What's wrong with this sound?"** — Call `get_sound_design_issues(track_index)` for a diagnostic without executing fixes
-- **"Show me the patch"** — Call `get_patch_model(track_index)` then `walk_device_tree(track_index)` for full device chain visibility
-- **"What can I automate?"** — Read the `controllable_params` list from the PatchModel response
-
-## Native Device Strengths
-
-When adding processing blocks, prefer native Ableton devices for controllability:
-
-- **Wavetable** — complex oscillator section with built-in modulation matrix
-- **Operator** — FM synthesis with per-operator envelopes and LFO
-- **Analog** — subtractive with two filters and two LFOs
-- **Auto Filter** — standalone filter with envelope follower and LFO
-- **Corpus** — resonant body modeling for physical character
-- **Erosion** — high-frequency noise and distortion artifacts
-- **Saturator** — waveshaping with multiple curve types
-
-Always `search_browser` before loading — never guess device names.
-
-## Deep Sound Design Reference
-
-Consult `references/sound-design-deep.md` for advanced techniques when working on creative requests. Key principles:
-
-### Making Sounds Breathe
-Every static sound can become alive with modulation below conscious perception:
-- **Filter breathing:** LFO at 0.1-0.5 Hz on filter cutoff, 5-15% depth
-- **Oscillator drift:** ±1-3 cent detune with very slow LFO (0.05-0.2 Hz)
-- **Amplitude micro-variation:** Perlin/brownian noise on volume, ±1-3 dB
-- **Rule:** If the listener can hear the modulation, it's too much. The best modulation is felt, not heard.
-
-### Space as Composition
-Reverb and delay are not decorations — in dub/minimal they ARE the composition:
-- **Dub chord:** Short stab → long delay (70-80% feedback) + filter on the delay return
-- **Delay throws:** Momentary send spikes (0→70% for half a beat) — the echo IS the event
-- **Sidechain reverb:** Dry drums trigger sidechain compression on reverb returns — the room pulses
-- **Feedback modulation:** Delay feedback at 75-85% + modulate delay time ±5-10% for warped echoes
-
-### Creative Sidechain (Beyond Pump)
-Sidechain compression is a modulation source, not just a mix tool:
-- **Sidechain filter:** Envelope follower from kick modulates pad filter cutoff — pad brightens between kicks
-- **Ghost sidechain:** Muted kick as sidechain source for textures — phantom groove on non-rhythmic elements
-- **Multiband sidechain:** Only duck sub frequencies from pad — shimmer stays, sub clears for kick
-
-### Effects as Instruments
-- **Self-oscillating filter:** Push resonance until it rings, play notes by changing cutoff
-- **Feedback loops:** Route output back to input through effects + compressor to control
-- **Convolution as synthesis:** Load non-IR files (speech, drum break) into convolution reverb — imprints spectral character
-- **Granular reverb:** Very short reverb (0.1-0.3s) high diffusion on percussion — smears transient into tonal cloud
-
-### The Frequency Dance
-At any moment, each frequency band should have one primary element. When one opens up, another pulls back:
-- Chord filter opens into highs → pull hi-hat back
-- Bass drops → kick shortens
-- Reverb tail fills → dry elements duck
-This is mix engineering as composition.
-
-### When to Apply These
-- User says "make it breathe" or "it sounds static" → micro-modulation
-- User says "more space" or "deeper" → dub techniques (delay throws, reverb composition)
-- User says "more groove" or "make it pump" → creative sidechain
-- User says "more texture" or "more complex" → textural layering
-- User says "surprise me" or "WTF moment" → brief textural disruption (2-8 beats max)
-- User says "warmer" or "more analog" → oscillator drift + subtle saturation + filter breathing

package/livepilot/skills/livepilot-sound-design-engine/references/patch-model.md

@@ -1,119 +0,0 @@
-# Patch Model Reference
-
-The PatchModel is returned by `get_patch_model(track_index)` and represents the complete signal processing structure of a track's device chain.
-
-## PatchModel Structure
-
-```json
-{
-  "track_index": 0,
-  "track_name": "Lead Synth",
-  "blocks": [
-    {
-      "device_index": 0,
-      "device_name": "Wavetable",
-      "block_type": "oscillator",
-      "controllable": true,
-      "parameters": ["Osc 1 Pos", "Osc 1 Gain", "Sub Gain", ...],
-      "sub_blocks": [
-        { "block_type": "filter", "parameters": ["Filter Freq", "Filter Res", ...] },
-        { "block_type": "lfo", "parameters": ["LFO 1 Rate", "LFO 1 Amount", ...] },
-        { "block_type": "envelope", "parameters": ["Amp Attack", "Amp Decay", ...] }
-      ]
-    },
-    {
-      "device_index": 1,
-      "device_name": "Auto Filter",
-      "block_type": "filter",
-      "controllable": true,
-      "parameters": ["Frequency", "Resonance", "Env. Modulation", ...]
-    }
-  ],
-  "controllable_params": [...],
-  "opaque_blocks": [],
-  "modulation_sources": [
-    { "type": "lfo", "device_index": 0, "rate": "0.5 Hz", "targets": ["Filter Freq"] }
-  ],
-  "signal_flow": "serial"
-}
-```
-
-## Block Types
-
-### oscillator
-Sound source generators. Found in synth instruments (Wavetable, Operator, Analog, Drift, Simpler, Sampler).
-
-Key parameters: waveform/position, pitch/tune/detune, unison voices/spread, sub oscillator level, FM amount/ratio.
-
-### filter
-Spectral shaping. Found as sub-blocks inside instruments or as standalone devices (Auto Filter, EQ Eight).
-
-Key parameters: cutoff frequency, resonance, filter type (LP/HP/BP/Notch), slope (12/24 dB), envelope modulation amount, drive.
-
-### envelope
-Time-shaping contour generators. Usually ADSR controlling amplitude or filter cutoff.
-
-Key parameters: attack, decay, sustain, release. Some instruments offer additional envelopes with arbitrary targets.
-
-### lfo
-Low-frequency oscillators for cyclic modulation.
-
-Key parameters: rate (Hz or synced), waveform (sine, triangle, square, saw, random), amount/depth, target parameter, phase, offset.
-
-### saturation
-Waveshaping, distortion, and harmonic generation.
-
-Devices: Saturator, Overdrive, Pedal, Amp, Erosion, Dynamic Tube. Key parameters: drive, tone, type/curve, dry/wet.
-
-### spatial
-Stereo field and space processing.
-
-Devices: Reverb, Delay, Chorus, Phaser, Flanger, Echo, Hybrid Reverb. Key parameters: decay/time, pre-delay, size, damping, dry/wet, feedback.
-
-### effect
-Catch-all for utility and dynamics processing.
-
-Devices: Compressor, Glue Compressor, Limiter, Gate, Utility, Tuner, Spectrum. Key parameters vary by device.
-
-## Controllable vs Opaque
-
-- **Controllable**: native Ableton devices and well-behaved plugins where `get_device_parameters` returns named, ranged parameters. All LivePilot set/get parameter tools work reliably.
-- **Opaque**: third-party AU/VST plugins where parameter inspection fails or returns generic names (Parameter 1, Parameter 2). The sound design engine can detect the block exists but cannot reason about individual parameters.
-
-Check `parameter_count` on opaque blocks. If <= 1, the plugin failed to load — delete it and replace with a native alternative.
-
-## Native Device Block Map
-
-Quick reference for which block types each native device provides:
-
-| Device | Primary Block | Sub-blocks |
-|--------|--------------|------------|
-| Wavetable | oscillator | filter, lfo, envelope |
-| Operator | oscillator | filter, lfo, envelope |
-| Analog | oscillator | filter, lfo, envelope |
-| Drift | oscillator | filter, lfo, envelope |
-| Simpler | oscillator | filter, lfo, envelope |
-| Sampler | oscillator | filter, lfo, envelope |
-| Auto Filter | filter | lfo, envelope |
-| EQ Eight | filter | — |
-| Compressor | effect | envelope (sidechain) |
-| Glue Compressor | effect | — |
-| Saturator | saturation | — |
-| Overdrive | saturation | — |
-| Pedal | saturation | — |
-| Reverb | spatial | — |
-| Delay | spatial | filter |
-| Echo | spatial | filter, lfo |
-| Chorus-Ensemble | spatial | — |
-| Phaser-Flanger | spatial | lfo |
-| Hybrid Reverb | spatial | — |
-| Corpus | spatial | — |
-| Utility | effect | — |
-
-## Rack Detection
-
-If a track contains an Instrument Rack or Audio Effect Rack, the PatchModel reports:
-
-- `signal_flow`: "parallel" (rack chains run in parallel)
-- Each chain appears as a nested block list
-- Use `get_rack_chains(track_index, device_index)` for per-chain detail

package/livepilot/skills/livepilot-sound-design-engine/references/sound-design-critics.md

@@ -1,118 +0,0 @@
-# Sound Design Critics Reference
-
-Each critic runs during `analyze_sound_design(track_index)` and produces an issue object with `critic`, `severity` (0.0-1.0), `block`, and `evidence`.
-
-## static_timbre
-
-Detects sounds that do not evolve over time. A static timbre lacks movement, making it sound flat and lifeless in a mix.
-
-**Thresholds:**
-- severity >= 0.7: No detectable spectral change over a 4-bar window. Crest factor of spectral flux < 0.5 dB
-- severity >= 0.4: Minimal change (spectral flux < 1.5 dB). Some subtle movement present
-- severity < 0.4: Adequate movement detected, likely intentional sustain character
-
-**Evidence format:**
-```json
-{
-  "spectral_flux_db": 0.3,
-  "analysis_window_bars": 4,
-  "brightest_moment_db": -8.2,
-  "dullest_moment_db": -8.5
-}
-```
-
-**Typical fix:** modulation_injection — add LFO to filter cutoff (rate 0.1-2 Hz, depth 20-40%).
-
-## no_modulation_sources
-
-Detects patches with zero modulation sources: no LFOs, no envelopes beyond amp/filter ADSR, no automation, no macro mappings.
-
-**Thresholds:**
-- severity >= 0.7: Zero modulation sources on a melodic/pad/lead instrument
-- severity >= 0.4: No modulation on a bass or rhythmic element (some genres keep bass static intentionally)
-- severity < 0.4: Drum hits and one-shots (modulation not expected)
-
-**Evidence format:**
-```json
-{
-  "lfo_count": 0,
-  "envelope_count": 1,
-  "automation_lanes": 0,
-  "macro_mappings": 0,
-  "instrument_type": "pad"
-}
-```
-
-**Typical fix:** modulation_injection — enable device LFO if available, or add Auto Filter with LFO. For pads, map LFO to filter cutoff and/or oscillator pitch (subtle detune).
-
-## modulation_flatness
-
-Detects modulation sources that exist but operate with such narrow ranges they produce no audible effect.
-
-**Thresholds:**
-- severity >= 0.7: Modulation depth < 5% of parameter range on all targets
-- severity >= 0.4: Modulation depth 5-15% — present but barely perceptible
-- severity < 0.4: Depth > 15%, audible modulation
-
-**Evidence format:**
-```json
-{
-  "modulation_source": "LFO 1",
-  "target_parameter": "Filter Cutoff",
-  "depth_percent": 3.2,
-  "parameter_range": [20, 20000],
-  "effective_range": [850, 920]
-}
-```
-
-**Typical fix:** Increase modulation depth to 20-40% for filter targets, 5-15% for pitch targets, 10-30% for amplitude targets.
-
-## missing_filter
-
-Detects raw oscillator output with no filter or spectral shaping in the signal path. Unfiltered signals often sound harsh and sit poorly in a mix.
-
-**Thresholds:**
-- severity >= 0.7: No filter device in the signal chain, and the source is a harmonically rich waveform (saw, square, FM)
-- severity >= 0.4: Filter present but fully open (cutoff at maximum, no resonance)
-- severity < 0.4: Filter active and shaping the spectrum
-
-**Evidence format:**
-```json
-{
-  "has_filter": false,
-  "source_waveform": "sawtooth",
-  "harmonic_content": "rich",
-  "device_chain": ["Wavetable"]
-}
-```
-
-**Typical fix:** filter_shaping — add Auto Filter (low-pass, cutoff around 2-5 kHz, resonance 20-40%) or enable the instrument's built-in filter.
-
-## spectral_imbalance
-
-Detects patches with too much energy concentrated in one frequency region, or significant spectral gaps.
-
-**Thresholds:**
-- severity >= 0.7: Single band > 12 dB above the mean, or a gap > 12 dB below
-- severity >= 0.4: Imbalance 6-12 dB in one or more regions
-- severity < 0.4: Balanced spectrum appropriate for the instrument role
-
-**Evidence format:**
-```json
-{
-  "peak_band": "high_mid",
-  "peak_deviation_db": 14.2,
-  "gap_band": "low_mid",
-  "gap_deviation_db": -8.5,
-  "band_levels_db": {
-    "sub": -22.0,
-    "low": -14.5,
-    "low_mid": -20.1,
-    "mid": -11.6,
-    "high_mid": -2.4,
-    "high": -8.9
-  }
-}
-```
-
-**Typical fix:** filter_shaping to tame the peak (EQ cut or filter sweep), or oscillator_tuning to fill the gap (add sub oscillator, adjust waveform).

package/livepilot/skills/livepilot-wonder/SKILL.md

@@ -1,79 +0,0 @@
----
-name: livepilot-wonder
-description: >
-  This skill should be used when the user asks to "surprise me",
-  "make it magical", "I'm stuck", "give me options", "take it somewhere",
-  or when stuckness confidence is high. Provides the Wonder Mode
-  stuck-rescue workflow with honest variant labeling and preview-first UX.
----
-
-# Wonder Mode — Stuck-Rescue Workflow
-
-Wonder Mode is a **preview-first stuck-rescue workflow**. It diagnoses
-why a session is stuck, generates genuinely distinct executable options,
-and lets the user preview, compare, and commit.
-
-## When to Trigger
-
-- User says "I'm stuck", "surprise me", "make it magical", "give me options"
-- `detect_stuckness` confidence > 0.5
-- 3+ consecutive undos in action ledger
-- Multiple plausible next moves with no clear winner
-
-## When NOT to Trigger
-
-- Exact operational requests ("set track 3 volume to -6dB")
-- Narrow deterministic edits ("quantize this clip")
-- Performance-safe-only context (unless explicitly requested)
-
-## The Workflow
-
-1. `enter_wonder_mode` — get diagnosis + 1-3 variants
-2. Explain the diagnosis in **musical language**, not tool language
-3. Present variants honestly:
-   - Executable variants: can be previewed and committed
-   - Analytical variants (`analytical_only: true`): directional ideas only
-4. `create_preview_set` with `wonder_session_id` for executable variants
-5. `render_preview_variant` for each executable variant
-6. `compare_preview_variants` — present recommendation with reasons
-7. User chooses:
-   - `commit_preview_variant` — applies the variant, records outcome
-   - `discard_wonder_session` — rejects all, keeps creative thread open
-
-## Honesty Rules
-
-- **Never describe an analytical variant as previewable**
-- **Never fabricate distinctness** by relabeling the same move
-- **Fewer than 3 variants is correct** when fewer distinct moves exist
-- 1 executable + 2 analytical is an honest, useful result
-- The `variant_count_actual` field tells you how many are real
-
-## Presenting Results
-
-For each variant, explain:
-- What it changes (in musical terms)
-- What it preserves (sacred elements)
-- Why it matters for this specific session
-- Whether it's executable or analytical-only
-
-For the recommendation, explain:
-- Why this one over the others
-- What risk it introduces
-- What sacred elements it preserves
-
-## Creative Intelligence (consult before generating variants)
-
-Wonder Mode should produce musically interesting results, not just technically correct ones. Before generating or applying any variant:
-
-1. Read `references/device-knowledge/automation-as-music.md` for automation shapes and macro gestures
-2. Read `references/device-knowledge/creative-thinking.md` for emotional-to-technical mapping
-3. Read `references/device-knowledge/chains-genre.md` if the session has a genre identity
-
-When reviewing Wonder variants, aim for musical depth:
-- **Filter arcs** — evolving filter across sections adds movement
-- **Space arcs** — reverb/delay sends breathing with density
-- **Micro-modulation** — subtle LFOs on sustained sounds
-- **Macro gestures** — coordinated multi-parameter moves at transitions
-
-Note: these are agent-level guidelines, not enforced by the Wonder engine.
-The engine generates variants from semantic moves; the agent adds musical polish.