Write a live-coded track as a Strudel pattern — mini-notation, synths, effects, song structure — and check it with the verdict.

MITAuto-check passed

Install Strudel

skills CLI
$ npx skills add autonomous-ai/openharness --skill strudel -a claude-code

Project install by default; add -g for ~/.claude/skills/.

GitHub CLI
$ gh skill install autonomous-ai/openharness strudel --agent claude-code

Project scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).

Manual copy
$ git clone --depth 1 https://github.com/autonomous-ai/openharness.git skills-src && mkdir -p .claude/skills && cp -r skills-src/store/agents/strudel/skills/strudel .claude/skills/strudel && rm -rf skills-src

Use ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.

Claude Code skills documentation · loads skills from .claude/skills/

Facts

Skill name
strudel
GitHub stars
1.1k
Token cost
~3k tokens
SKILL.md length
1,476 words
Files
1
Skills in repo
100
Repo updated
First seen
Licence
MIT

At a glance

Write a live-coded track as a Strudel pattern — mini-notation, synths, effects, song structure — and check it with the verdict.

  • Works in 5 steps: Write track.strudel with a tempo and one… → Build up a voice at a time, running the… → Prefer synths. If the user asks for real… → …
  • The workspace holds a .strudel file
  • SKILL.md covers Offline first: use synths, not…, Mini-notation, Notes, scales, chords and Shaping the sound, plus 5 more sections
  • Calls python3

What it does

Strudel is an agent skill from autonomous-ai/openharness. Write a live-coded track as a Strudel pattern — mini-notation, synths, effects, song structure — and check it with the verdict. Use whenever the workspace holds a .strudel file or the user asks for a beat, a groove, a bassline, a loop, a track, or music in the pane.

Its SKILL.md is about 3k tokens, which your agent loads only when the skill is triggered. It is a single SKILL.md file with no bundled scripts.

It works with JavaScript. The repository describes itself as: The ultimate harness for coding agents and beyond. All your agents. All your machines. One command center. Start with code, then follow your curiosity and build across… The licence is MIT.

When your agent uses it

  • The workspace holds a .strudel file
  • The user asks for a beat
  • Music in the pane

Example prompts

  • “/strudel”

Requirements

  • Python 3

Workflow steps

5 steps, taken from the first numbered list in SKILL.md.

  1. Write track.strudel with a tempo and one or two voices within the first minute, run the
  2. Build up a voice at a time, running the verdict after each save. A file that does not parse
  3. Prefer synths. If the user asks for real drums, use them and say the pane needs the internet.
  4. Ask only what you cannot infer — tempo, key, mood, length. Otherwise decide, say so in one line,
  5. Deliver track.strudel. It is plain text and pastes straight into strudel.cc.

What it can do on your machine

Read from SKILL.md and the folder at commit 50da5db. It shows what the files ask for, not the result of running them.

  • Tool permissions

    Pre-approves nothing: there is no allowed-tools line, so your agent's usual permission prompts apply.

    From allowed-tools in the SKILL.md frontmatter.

  • Runs code

    Shell commands in SKILL.md call:

    • python3

    From the folder's file list and the shell code blocks in SKILL.md.

  • Network

    No URLs in SKILL.md.

    From URLs in SKILL.md, links to its own repository left out.

  • Credentials

    Names no API keys, tokens, secrets or passwords.

    From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.

Context cost

Strudel loads about 3k tokens when it runs. Until then it costs about 69 tokens; SKILL.md has 1,476 words of instructions outside code blocks.

Always · name and description, kept in context so the agent knows when to use it
~69
When it runs · the whole SKILL.md, loaded when a task matches
~3k

Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.

Safety

Auto-check passed

The automated check found no risky patterns in SKILL.md.

Automated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); files beside SKILL.md are not scanned.

SKILL.md

The full file from autonomous-ai/openharness at commit 50da5db, republished under its MIT licence (© autonomous-ai). 1,476 words, ~2,964 tokens.

Download SKILL.mdSave it as .claude/skills/strudel/SKILL.md (or your agent's skills folder).
name
strudel
description
Write a live-coded track as a Strudel pattern — mini-notation, synths, effects, song structure — and check it with the verdict. Use whenever the workspace holds a .strudel file or the user asks for a beat, a groove, a bassline, a loop, a track, or music in the pane.

Strudel

Strudel is TidalCycles in JavaScript. A track is one expression that describes a cycle; the REPL plays that cycle forever until you change it. You write one file, track.strudel, and the pane plays it — a save hot-swaps the pattern on the next cycle without stopping the transport.

The whole file is evaluated as ECMAScript 2022 (top-level await allowed), so JavaScript syntax rules apply: balanced brackets, balanced quotes, // and /* */ comments. There is no module loader — never write import or require. Everything Strudel offers is already in scope.

Check after every save:

sh
python3 "$STRUDEL_TOOLCHAIN/verdict.py"          # track.strudel
python3 "$STRUDEL_TOOLCHAIN/verdict.py" b.strudel

It parses the file with the same parser the REPL uses, checks offline sound sources, and writes .harness/verdict.json, which is what the pane header shows. It does not listen to the music. Playing is the pane's job; musical quality needs listening and judgment.

Offline first: use synths, not samples

The REPL preloads sample maps from GitHub at start and fetches each .wav the first time it is triggered. So s("bd sd hh") needs the internet, twice. These sound sources need nothing:

SourceNames
Oscillatorssine sawtooth square triangle (aliases sin saw sqr tri)
Richersupersaw (unison, spread, detune), pulse (pw, pwrate, pwsweep)
Noisewhite pink brown (hard → soft), crackle (with density)
Drumsbd, a synthesised bass drum
Chipzzfx z_sine z_sawtooth z_triangle z_square z_tan z_noise

Build drums out of those: s("sbd*4") is the kick, short noise is the snare and the hats.

js
stack(
  s("sbd*4").decay(0.24).lpf(280),
  s("~ white ~ white").decay(0.11).sustain(0).hpf(1200).gain(0.5),
  s("white*8").decay(0.03).sustain(0).hpf(7000).gain("[0.3 0.14]*4")
)

Reach for bd sd hh oh rim cp and .bank("RolandTR909") only when the user wants real drum machines and has a connection. Say so when you do; the verdict warns about it too.

If you set note() and no s(), the sound defaults to triangle — already offline.

Mini-notation

Everything inside a double-quoted string is mini-notation. Single quotes are a plain string, not a pattern.

a b ca sequence; the whole thing still takes one cycle, so more events = shorter events
~ or -a rest
[a b]a subdivision — one slot of the outer sequence, nests as deep as you like
<a b c>one per cycle. <a b c> is exactly [a b c]/3
a*2faster / repeat within the slot (decimals allowed: *2.75)
a/2slower — spread over 2 cycles
a, bin parallel (a stack), at the top level or inside [ ]
a!3replicate: a a a, without speeding up. Bare ! means 2
a@3elongate: this one gets 3 slots' worth of time. Default weight 1
a? a?0.1drop it 50% / 10% of the time
a|bpick one at random each cycle
a(3,8) a(3,8,1)Euclid: pulses, steps, and an optional rotation
{a b c, x y} {a b c}%4polymeter
bd:2second value — for s() that is the sample index, a third is gain
js
s("bd [hh hh] sd [hh bd]")                      // subdivision
note("<c2!2 g1!2 as1!2 f1!2>(3,8)").s("sawtooth") // 8 bars of euclidean bass
s("[bd sd]*2, white*8")                          // two voices in one string
n("<4 [3@3 4] [<2 0> ~] ~>").scale("d4:minor")   // weights and nesting

Notes, scales, chords

  • note("c3 e3 g3") — letters a–g, b/#, optional octave 0–9, default octave 3. MIDI numbers work too (note("48 55"), 69 = A4).
  • n("0 2 4").scale("c4:minor") — degrees of a scale, zero-indexed, negatives wrap backwards. The scale name takes no spaces: "c:minor", "f#4:mixolydian", "a:harmonic:minor".
  • [c3,eb3,g3] inside a pattern is a chord — a stack of notes in one slot.
  • n("0 1 2 3").chord("<C Am F G>").voicing() builds voicings from chord symbols.
  • .arp("0 [0,2] 1 [0,2]") spreads a stacked chord over time. It takes a pattern of indices, not mode names like "up".
js
note("<[c3,ds3,g3] [g2,as2,d3] [as2,d3,f3] [f2,gs2,c3]>")
  .s("supersaw").unison(5).spread(0.45).attack(0.5).release(1.2).room(0.6)

Shaping the sound

  • Envelope attack/att, decay/dec, sustain/sus, release/rel, or all four at once as .adsr(".1:.1:.5:.2"). clip (alias legato) multiplies note length.
  • Filter lpf (aliases cutoff, lp), hpf, bpf; resonance is lpq (alias resonance, 0–50), hpq, bpq. ftype picks 12db / ladder / 24db.
  • Filter envelope lpattack lpdecay lpsustain lprelease lpenv (short: lpa lpd lps lpr lpe). lpenv is the depth and may be negative.
  • Space room (0–1) with roomsize (alias rsize/size, 0–10 — change it sparsely); delay, delaytime/dt, delayfeedback/dfb. orbit(n) gives a voice its own reverb and delay bus; patterns on the same orbit share one.
  • Colour distort (prefer it over the deprecated shape), crush (1 crunchy … 16 clean), coarse, vowel("<a e i o u>"), phaser.
  • Level and place gain, postgain, pan (0 left … 1 right).

Many of these take several values through : — lpf("1000:10") is cutoff and resonance, delay("0.65:0.25:0.9") is level, time and feedback, room("0.9:4") is level and size.

js
note("c3 bb2 f3 eb3").s("sawtooth").lpf(600).adsr(".1:.1:.5:.2")
note("c2 e2 f2 g2").s("sawtooth").lpf(300).lpa(0.5).lpenv("<4 2 1 0 -1 -2 -4>/4")

Moving the pattern

rev palindrome iter(4) chunk(4, f) ply(2) off(1/16, f) superimpose(f) layer(f) jux(rev) juxBy(0.5, f) segment(16) struct("x ~ x x") mask("1 0") degradeBy(0.3) sometimes(f) sometimesBy(0.4, f) often (0.75) rarely (0.25) almostAlways (0.9) someCyclesBy(p, f) every(4, f) slow(2) fast(2) euclid(3,8) euclidRot(3,8,1) euclidLegato(3,8) add(note("12")) echo(3, 1/8, 0.5) linger swingBy(1/3, 4) hurry(2).

js
n("0 1 [4 3] 2 0 2 [~ 3] 4").s("triangle").jux(rev)
s("white*8").sometimesBy(0.4, x => x.hpf(2000).gain(0.4))
n("0 [4 <3 2>] <2 3> [~ 1]".off(1/16, x => x.add(4))).scale("c5:minor").s("triangle").dec(0.1)

Signals as modulators

sine cosine saw isaw tri square rand perlin run 0…1 (the …2 variants, sine2 saw2 rand2, run −1…1); irand(n) is called as a function. Shape them with .range(lo, hi) (or .rangex for exponential) and slow them with .slow(n).

js
note("<[c2 c3]*4 [bb1 bb2]*4>").s("sawtooth").lpf(sine.range(100, 2000).slow(4))
n(saw.range(0, 8).segment(8)).scale("c:major")
n(irand(8)).struct("x x*2 x x*3").scale("c:minor")

The catch worth knowing: a signal is sampled once per event, not continuously. A pad held for a whole cycle gets one filter value per note, not a sweep. Add events to hear the movement:

js
s("supersaw").seg(16).lpf(tri.range(100, 5000).slow(2))

Envelopes are continuous in time: attack/decay/release, the pitch envelope (penv, pattack, pdecay), the filter envelopes, tremolo, phaser, vib.

Show full SKILL.md (651 more words)Show less

Putting a track together

Tempo. setcpm(120/4) — cycles per minute, and with one bar per cycle that is 120 bpm in 4/4. setcps(x) is the same thing per second; the default is 0.5 cps, a 2-second cycle. Set it on the first line so the file says its own tempo.

Layers. stack(a, b, c) plays voices together and is the shape to reach for. The $: label form does the same thing one line at a time, and _$: mutes a line — handy when the user asks to drop a part:

js
setcpm(120/4)
$: s("sbd*4").decay(0.24)
$: note("<c2 g1>(3,8)").s("sawtooth").lpf(800)
_$: n("0 3 7").scale("c4:minor").s("triangle")   // muted

Name every voice. The pane draws one lane per stack argument (or $: line) — its notes scrolling under a playhead, an activity light, Mute and Solo — and titles the lane from the comment directly above the voice, up to the first —, : or .. So give every voice its own argument and a title line, and keep what it does after the dash:

js
stack(
  // Kick — four on the floor
  s("sbd*4").decay(0.24),
  // Bass — three pulses in eight, filter opening over four bars
  note("<c2 g1>(3,8)").s("sawtooth").lpf(sine.range(220, 1600).slow(4))
)

A named label (bass: note(...)) is its own title. Without a comment the lane is named after the first sound it plays.

Structure. A cycle is a bar. Longer form comes from <> (one per cycle), ! and @ to hold a chord for two bars, /n to stretch, and every(n, f) / someCyclesBy to vary. Eight bars of harmony is <[chord]!2 [chord]!2 [chord]!2 [chord]!2>.

  • Intro — one or two voices, filter closed, no lead. .lpf(400) on everything is a whole intro.
  • Groove — kick, backbeat, hats, bass, then harmony. Give each voice its own register and one clear rhythmic idea; do not let two voices play the same subdivision at the same volume.
  • Break — .mask("<1 1 1 0>") drops a voice for the fourth bar of four; every(4, x => x.fast(2)) doubles it instead. A break is one voice leaving, not everything changing.
  • Lift — open the filter, add .jux(rev), add an octave with .sometimes(add(note("12"))).

Mix. Keep gain roughly: kick 0.9–1, snare 0.5, hats 0.15–0.3, bass 0.8, pad 0.3, lead 0.3. Reverb on one or two voices only, and roomsize set once. Everything at 1 is mud.

Effects are single-use per voice. s("bd").lpf(100).distort(2).lpf(800) does not chain — the last lpf wins. Use superimpose or a second stack entry for a second treatment.

How to work

  1. Write track.strudel with a tempo and one or two voices within the first minute, run the verdict, and tell the user to click Play. The pane hot-swaps from then on.
  2. Build up a voice at a time, running the verdict after each save. A file that does not parse leaves the pane playing the last good pattern and shows the error, with its line, above the code — the music never breaks, but nothing new is heard until the verdict is clean.
  3. Prefer synths. If the user asks for real drums, use them and say the pane needs the internet.
  4. Ask only what you cannot infer — tempo, key, mood, length. Otherwise decide, say so in one line, and play it.
  5. Deliver track.strudel. It is plain text and pastes straight into strudel.cc.

Develop a performance the user kept

Record take captures the actual stereo mix while the user performs: Mute/Solo, Ctrl+Enter code changes and named moments. Finish take → Keep take preserves the WAV, source versions and take.json journal in out/takes/<id>/; Audio + source downloads them as a ZIP. This works for any pattern the engine plays, including original synth-only tracks.

If the user names a take or moment, inspect that journal first. Find the last source event before the marker, read its numbered .strudel file and account for the mute/solo requests. Continue composing from that intention in the live track. Leave the kept take intact. Event times describe control requests; scheduled notes, effect tails, randomness and external dependencies mean the journal is not an exact audio replay. The recorded WAV is the performance to compare against.

File updates wait while the user records or has unsaved edits. The performer applies them with Load new version. Mention this only when relevant to their next revision.

© autonomous-ai, MIT. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file

Files

Just SKILL.md in store/agents/strudel/skills/strudel of autonomous-ai/openharness.

Open the folder on GitHubat commit 50da5db

Compare with similar skills

Strudel next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.

Strudel compared with similar skills
SkillStarsUsed inTokensAuto-checkLicenceRepo updated
Strudel this skillautonomous-ai/openharness1.1k—~3kAutomated safety check: PassMIT
Code Review ChecklistshareAI-lab/learn-claude-code78k5 repos~1.1kAutomated safety check: PassMIT
Tailwindcss Developmentanonaddy/anonaddy4.9k10 repos~865Automated safety check: PassMIT
Figma use_figma Plugin API Ruleswarpdotdev/warp65k4 repos~4.4kAutomated safety check: PassAGPL-3.0
Install Anti-Slop Oxlint Rulesdmmulroy/anti-slop5.3k1 repos~2.2kAutomated safety check: PassMIT
GSAP Core Animationgreensock/gsap-skills16k4 repos~3.7kAutomated safety check: PassMIT

Similar skills

  • Code Review Checklist

    shareAI-lab/learn-claude-code

    Reviews code against a five-part checklist covering security, correctness, performance, maintainability and testing, and reports findings in a fixed format.

    78k GitHub starsUsed in 5 repos~1.1k tokens
    DevelopmentAuto-check passed
  • Tailwindcss Development

    anonaddy/anonaddy

    Always invoke when the user's message includes 'tailwind' in any form.

    4.9k GitHub starsUsed in 10 repos~865 tokens
    Frontend & DesignAuto-check passed
  • Required groundwork before any use_figma call: the rules and reference files for running JavaScript in a Figma file through the Plugin API without common failures.

    65k GitHub starsUsed in 4 repos~4.4k tokens
    Frontend & DesignAuto-check passed
  • Installs, updates or migrates the vendored anti-slop Oxlint plugin in a repository, keeping local rule changes and the plugin's license and provenance files.

    5.3k GitHub starsUsed in 1 repo~2.2k tokens
    DevelopmentAuto-check passed
  • GSAP Core Animation

    greensock/gsap-skills

    Covers the GSAP core API for tweens, easing, staggers, defaults and matchMedia, and when to choose GSAP over CSS animations or other JavaScript animation libraries.

    16k GitHub starsUsed in 4 repos~3.7k tokens
    Frontend & DesignAuto-check passed
  • Chrome CDP Browser Control

    zenstory-ai/oh-story-claudecode

    Drives a Chrome window over the DevTools Protocol with the agent-browser CLI, so the agent can reuse your logged-in sessions, read pages and pull tokens.

    7.4k GitHub starsUsed in 3 repos~1.2k tokens
    Productivity & AutomationAuto-check passed

More from autonomous-ai/openharness

All 100 skills in this repo
  • G-code Slicer Tool

    autonomous-ai/openharness

    Slices 3D mesh files into printer-profiled plain G-code through real slicer CLIs, with backend discovery, input inspection, dry runs and static validation.

    1.1k GitHub stars~1.5k tokensUpdated today
    Auto-check passed
  • Home Assistant Automation Builder

    autonomous-ai/openharness

    Turns a home-automation request into standard, testable automations.yaml, run against Home Assistant Core's real triggers and verified with its own trace tool.

    1.1k GitHub stars~1.2k tokensUpdated today
    Auto-check passed
  • Score Music Composer

    autonomous-ai/openharness

    Turns a musical brief into LilyPond concert-pitch music, checked parts for each instrument and a playable practice pack.

    1.1k GitHub stars~1.1k tokensUpdated today
    Auto-check passed
  • OrcaSlicer 3MF and G-code Workflow

    autonomous-ai/openharness

    Turns an STL and explicit printer and material requirements into compared OrcaSlicer plans, an editable 3MF project, checked G-code and a portable handoff.

    1.1k GitHub stars~1.5k tokensUpdated today
    Auto-check passed
  • Sheets and Docs Report Builder

    autonomous-ai/openharness

    Builds an editable DOCX report, a formula-driven XLSX workbook and a fresh LibreOffice PDF preview from one structured source file, then checks them together.

    1.1k GitHub stars~708 tokensUpdated today
    Auto-check passed
  • Bambu Labs

    autonomous-ai/openharness

    Dry-run, upload, and cautiously initiate local Bambu Lab print jobs from validated plain .gcode, using Bambu LAN FTPS/MQTT handoffs.

    1.1k GitHub starsUsed in 1 repo~2.1k tokens
    Auto-check: warnings

Works with

Questions about Strudel

What does Strudel do?

Write a live-coded track as a Strudel pattern — mini-notation, synths, effects, song structure — and check it with the verdict. Strudel is an agent skill from autonomous-ai/openharness. Write a live-coded track as a Strudel pattern — mini-notation, synths, effects, song structure — and check it with the verdict.

When should I use Strudel?

Strudel fits situations like: the workspace holds a .strudel file; the user asks for a beat; music in the pane.

How do I install Strudel in Claude Code?

Run `npx skills add autonomous-ai/openharness --skill strudel -a claude-code`. Or copy the skill folder (store/agents/strudel/skills/strudel in autonomous-ai/openharness) into .claude/skills/strudel in your project. Claude Code loads it when a task matches its description.

How do I install Strudel in Codex?

Run `npx skills add autonomous-ai/openharness --skill strudel -a codex`. Or copy the skill folder (store/agents/strudel/skills/strudel in autonomous-ai/openharness) into .agents/skills/strudel in your project. Codex loads it when a task matches its description.

Can I use Strudel in Cursor, Gemini CLI or GitHub Copilot?

Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add autonomous-ai/openharness --skill strudel -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/strudel, .gemini/skills/strudel, .github/skills/strudel and .opencode/skills/strudel in your project.

What does Strudel need to run?

Going by SKILL.md and its folder, Strudel needs the command-line tools its instructions call (python3). Our summary lists: Python 3.

Does Strudel access the network?

SKILL.md contains no URLs. Any network use would come from the scripts or tools the agent runs. This is read from the text; nothing was executed.

Is Strudel safe to install?

Our automated static check of SKILL.md found no risky patterns, such as piping downloads into a shell, reading credential files or hidden Unicode. It is not a guarantee. Review the folder before installing.

What licence does Strudel use?

Strudel is published under the MIT licence (the repository's licence). It allows redistribution, so the full SKILL.md is shown on this page.

How many tokens does Strudel use?

About 3k tokens (SKILL.md is roughly 12k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full.

What are the alternatives to Strudel?

Skills that share tags, products or a category with Strudel: Code Review Checklist (shareAI-lab/learn-claude-code, 78k stars), Tailwindcss Development (anonaddy/anonaddy, 4.9k stars), Figma use_figma Plugin API Rules (warpdotdev/warp, 65k stars) and Install Anti-Slop Oxlint Rules (dmmulroy/anti-slop, 5.3k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.

Who maintains Strudel?

autonomous-ai (a GitHub organization) maintains it in autonomous-ai/openharness, which has 1,149 GitHub stars. The repository holds 100 skills in this directory. The repository was last updated on October 8, 2026.

Source: autonomous-ai/openharness on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.