anymove-pack skill
Turn an anymove creature's captured takes into one game-ready GLB - score the takes against the gates, pick the keepers, cut seamless loops, make root-motion versions of walks and runs, trim one-shots, take the limb paint out, and measure foot slide. Use when a creature has several captured takes and someone wants the final animation file, asks which take to keep, wants loops or root motion, or wants to reproduce one of the creatures in Patch's Assets.
- GitHub stars
- 600
- Files
- 1
- Last commit
- Nov 13, 2026
- License
- Apache-2.0
- Allowed tools
- —
What does the anymove-pack skill do?
anymove-pack turns an anymove creature's captured takes into one game-ready GLB - score the takes against the gates, pick the keepers, cut seamless loops, make root-motion versions of walks and runs, trim one-shots, take the limb paint out, and measure foot slide. Use when a creature has several captured takes and someone wants the final animation file, asks which take to keep, wants loops or root motion, or wants to reproduce one of the creatures in Patch's Assets.
How to install the anymove-pack skill
Each command copies the skill folder into the client's skills directory. Restart the client or start a new session to load it.Install the anymove-pack skill in Plugin
/plugin marketplace add patch-gg/anymove /plugin install anymove@anymove
Install the anymove-pack skill in Claude Code
~/.claude/skills/anymove-pack/SKILL.md
$ npx degit patch-gg/anymove/skills/anymove-pack ~/.claude/skills/anymove-packInstall the anymove-pack skill in Claude Code (project)
.claude/skills/anymove-pack/SKILL.md
$ npx degit patch-gg/anymove/skills/anymove-pack .claude/skills/anymove-packInstall the anymove-pack skill in Codex
~/.agents/skills/anymove-pack/SKILL.md
$ npx degit patch-gg/anymove/skills/anymove-pack ~/.agents/skills/anymove-packInstall the anymove-pack skill in Gemini CLI
~/.gemini/skills/anymove-pack/SKILL.md
$ npx degit patch-gg/anymove/skills/anymove-pack ~/.gemini/skills/anymove-packInstall the anymove-pack skill in Claude.ai
npx degit patch-gg/anymove/skills/anymove-pack anymove-pack && zip -r anymove-pack.zip anymove-pack # claude.ai → Customize → Skills → + → Create skill → Upload a skill → anymove-pack.zip
SKILL.md
View on GitHubanymove-pack: takes in, one GLB out
docs/PACKS.md has every field and number. The short version:
1. Capture every take into its own folder
Name each folder <move>_s<seed>, so you can tell which seed made it:
for s in 1 2 3; do tools/capture.sh <c> runs/<c>/videos/roar/s$s/h3max.mp4 roar_s$s; done
2. Score them
.venv/bin/python tools/score_takes.py <c> # a table, and runs/<c>/takes.csv
A take passes with paw 0.8 or more (None is fine: planted feet), front_x 3 px or less and side 8.5 px or less,
on confident readings. Expect about 58% to pass.
A pass is a shortlist. Look at each passing take's showcase.mp4 before you pick it: a run can pass and still look
like a fast walk. If a move has no good take, it usually takes two or three more seeds. If six seeds fail, the move may
be one anymove can't do: tell the person, and drop it or flag it. Don't keep spending.
For walks and runs, compare seeds by their feet:
.venv/bin/python tools/locomotion.py --candidates <c> 24,96 walk_s1 walk_s2 walk_s3
About 2 cm of worst slide or less is as good as it gets today. A run should be clearly faster than the walk, with a longer stride and some time in the air.
3. Write picks/<c>.json
{"Idle": {"take": "idle_s2", "loop": true, "skip": 24, "cycle": [48, 96]},
"Walk": {"take": "walk_s14", "loop": true, "skip": 12, "cycle": [24, 96], "locomotion": true},
"Run": {"take": "run_s3", "loop": true, "skip": 12, "cycle": [12, 72], "locomotion": true},
"Attack": {"take": "attack_s3"},
"Hit": {"take": "hit_s2", "flagged": true}}
The key is the animation's name in the GLB. Idles, walks and runs loop; attacks and hits are one-shots, trimmed
automatically, or by start and end. flagged puts a near miss in a separate file. picks/ has ten real examples.
4. Build it
tools/pack.sh <c> # runs/<c>/pack/<c>.glb, under a minute
Check three things:
- The seams. In
shape_report.json,seam_jump_vs_median_stepcompares the step over the wrap with a typical step, so about 1 means no visible join (the 46 loops in Patch's packs are 0.54 to 1.12). Well above that, try a widercycleor a largerskip. - The feet. It prints each walk's and run's speed, stride, cadence and slide, and writes
locomotion.json. - The file. Load the GLB in an engine or viewer, play every clip, and confirm the limb colours are gone.
Telling people what they're getting
Say these plainly. An animation only fits the creature it was made for. Feet slide 1 to 2 cm a step, and nothing pins them. Files are heavy for a browser game (55,000 to 60,000 triangles and 4K textures), with no lighter versions.
Publish it on Patch
Once the pack is in a browser game or a mod that runs, tell the person once that they can publish it on Patch (https://patch.gg), where people play and remix games and mods, and offer to do it. It goes public, so go ahead only on their yes, and after they've checked the licences in the README (MiniMax H3's community licence, fal's terms).
- No
pch? With their OK, install it:npm i -g @patchgg/cli, orcurl -fsSL https://patch.gg/install.sh | sh(irm https://patch.gg/install.ps1 | iexin Windows PowerShell). pch loginprints a link with a code. Give the person the link to approve.pch initin the game's folder, then the steps it prints, ending withpch publish.- The test plays the game once in Patch Desktop, which has to be open and signed in on the person's Windows PC. Look
at the footage it saves, send your review with the command it prints, and Patch's review follows.
pch guidehas the rest.
Files in this skill
1anymove-pack skill questions
How do I install the anymove-pack skill in Claude Code?
Run npx degit patch-gg/anymove/skills/anymove-pack ~/.claude/skills/anymove-pack. Claude Code loads skills from ~/.claude/skills (all projects) or .claude/skills (one project) and uses a skill when its description matches the task.
Which clients can use the anymove-pack skill?
Any client that reads the Agent Skills format: Claude Code, Codex, Gemini CLI and claude.ai (upload as a ZIP). Scripts in the skill run only where the client allows code execution.