agentleFS
glinckerglincker/ableton-skills/skills/midi-cleanup/SKILL.md

midi-cleanup

Skill32 starsChanged 55 days agoMITView on GitHub

Use when the user asks to humanize a MIDI part, fix voice leading, quantize notes, clean up timing, fix stuck/duplicate notes, or generally polish a MIDI clip they recorded or wrote. Examples - "humanize this hat", "voice these chords better", "the timing is too stiff", "fix the voice leading on the strings".

The file

We split this skill into the rules it gives an agent

10 rules. About 0 are common, 9 are distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈9

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  • could be automatic: a setting could enforce it
  1. Distinctive:

    Default range: ±10-15 velocity for hats/percussion, ±8-12 for drums

    line 27
  2. Distinctive:

    Add micro-timing offsets: ±5-15 ticks (5-15 ms at 120 BPM) with a slight forward bias on off-beats for groove

    line 29
  3. Distinctive:

    Don't humanize the downbeat of bar 1 — it's the anchor

    line 38
  4. Distinctive:

    Keep the bass note in the bass voice

    line 45
  5. Distinctive:

    Avoid parallel fifths and octaves between outer voices (counterpoint baseline)

    line 49
  6. Distinctive:

    Don't auto-quantize 100% — that destroys feel.

    line 55
  7. Distinctive:

    Default strength: 75% to 16th-note grid

    line 57
  8. Very specific:

    Use batch operations if the MCP server supports them (set_clip_notes with the full new note array is preferred over per-note edits — …

    line 76
  9. Distinctive:

    Don't merge a clip you just modified: Always leave undo on the table.

    line 91
  10. Distinctive:

    Read the hat clip — 32 16th notes, all velocity 100, perfectly on grid.

    line 98

≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

MIDI Cleanup

You are polishing a MIDI clip the user wrote or recorded. Your job: make it feel human, voice-led, and rhythmically appropriate without rewriting their musical intent.

Workflow

1. Identify the target clip

User must specify which clip. If ambiguous, ask:

"Which clip — the lead on track 3 or the chords on track 5?"

Then read the clip with get_notes_from_clip (or equivalent). Note the note count, range, and rhythmic density.

2. Identify the cleanup type

Match the user's request to one of these operations:

A. Humanize velocity

For drums, hats, percussion, or any rhythmic part that feels stiff.

  • Default range: ±10-15 velocity for hats/percussion, ±8-12 for drums
  • Preserve accents: notes already at velocity >100 stay loud; notes <40 stay quiet
  • Add micro-timing offsets: ±5-15 ticks (5-15 ms at 120 BPM) with a slight forward bias on off-beats for groove
  • For shuffled feels: add 2-5% swing on 16th notes

B. Humanize timing

For piano, guitar, or string parts that sound MIDI-stiff.

  • ±8-20 ms timing variance, slightly more on off-beats
  • Chord notes: stagger attack 5-15 ms (top note slightly later) for "rolled" feel — only if the user wants that
  • Don't humanize the downbeat of bar 1 — it's the anchor

C. Voice leading fix

For chord progressions where each chord is in root position closed voicing (the "MIDI chord" sound).

Rules:

  • Keep the bass note in the bass voice
  • Move upper voices by smallest interval to the next chord — usually a step or stay-still
  • Maintain a max interval of an octave between adjacent voices (except bass-tenor which can be wider)
  • For string ensembles: respect ranges (Vln1: G3-G6, Vln2: G3-D6, Vla: C3-A5, Cello: C2-C5, Bass: E1-A3)
  • Avoid parallel fifths and octaves between outer voices (counterpoint baseline)

Apply by reading the current chord notes, computing the optimal voicing for each chord, and rewriting.

D. Quantize (intelligent)

Don't auto-quantize 100% — that destroys feel.

  • Default strength: 75% to 16th-note grid
  • For shuffled genres (hip-hop, lo-fi): 60-70% to a swing grid
  • For tight EDM/house: 90-100% on percussion only, looser on lead/chords
  • Preserve grace notes and intentional pickups (notes within 30 ms of another note ahead of grid)

E. Stuck/duplicate note fix

  • Find notes with duration < 5 ms → delete
  • Find overlapping notes with same pitch → merge or trim to non-overlap
  • Find sustained notes longer than 2 bars in fast parts → suspect, ask user

3. Preview the changes

Before writing, summarize:

"I'll humanize velocity ±12 on the 32 hat notes (preserving the 4 accent hits at velocity 110), and add ±8 ms timing variance with a slight late-bias on the off-beats. Apply?"

4. Apply

Use batch operations if the MCP server supports them (set_clip_notes with the full new note array is preferred over per-note edits — atomic, undoable in one Ctrl+Z).

5. Confirm result

  • State what changed
  • Suggest the user listen and tell you if they want more/less of the effect
  • Offer to revert with one command if it doesn't feel right

Don'ts

  • Don't change pitches unless doing voice leading — and even then, never the melody/lead voice.
  • Don't humanize beyond ±25 velocity or ±25 ms — that becomes noticeable as "bad timing" rather than "human."
  • Don't quantize at 100% unless explicitly requested. Loss of feel is the #1 complaint about MIDI.
  • Don't apply voice leading to a melody. Voice leading is for chords/pads/inner voices, not the lead line.
  • Don't touch the velocity 127 hits — those are intentional accents.
  • Don't merge a clip you just modified. Always leave undo on the table.

Example

User: "The hi-hat sounds too robotic, humanize it"

Steps:

  1. Read the hat clip — 32 16th notes, all velocity 100, perfectly on grid.
  2. Plan: ±12 velocity (preserve any accents), ±8 ms timing with off-beat late-bias of +3 ms, slight 4% swing.
  3. Preview: "32 notes, no existing accents detected. I'll spread velocity 88-112, add 8 ms timing variance, 4% swing on off-beats. Apply?"
  4. On confirm, rewrite the clip atomically.
  5. Report: "Done. Listen — if it's still too tight, say 'more humanize' and I'll go to ±18 velocity."

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.