YouTube MCP for Claude & ChatGPT
TubeLab is the YouTube MCP Server for Claude and ChatGPT. Real channels, viral outliers, video transcripts and much more - all in your AI chat.
翻訳待ち
The maker's introduction consists of the line above. Read the repository README for more details. Repository README ↗
Loading the full README.
Read the full README ↗One repository holding the Claude Code skills used across every project, plus
wska — the CLI that links a chosen subset into a project or into the user's
global ~/.claude/.
It replaces batleforc/weebo-skills (now deleted),
which published skills as a plugin marketplace and is abandoned.
A skill is a plain directory containing a SKILL.md. wska install symlinks
the directories named by a profile into the target's .claude/skills/, so
editing a skill here is live in every target immediately, with no reinstall
step.
Two items cannot be symlinked, because each writes state coupled to the project
it is installed in: bmad renders per-project snapshots from an
installer-generated _bmad/config.toml, and graphify is installed by pipx and
writes its own skill directory. Both are declared as installers in
vendor.toml, and wska shells out to them instead of linking.
Requires the toolchain from mise.toml (task mise:install), which brings
Rust, Task, cocogitto and most of what the items themselves need: node, uv,
graphify, gitleaks, cargo-audit.
task init # full bootstrap: mise, plugins, hooks, all items into this repo
task skills:install PROFILE=design # one profile into this repo's .claude/, recorded in .claude/wska.toml
task skills:add ITEM=weebo-techdoc # add one skill to what .claude/wska.toml already records
task skills:remove ITEM=weebo-techdoc # and drop it again
task skills:install PROFILE=frontend SCOPE=user # the same into ~/.claude/
task skills:update # git pull --ff-only the items repo, then re-apply .claude/wska.toml
task skills:status # what is linked where
task skills:doctor # can each item actually work on this machine
To use the skills from another project, install the released binary and run
wska install there — it clones this repo on first use; see
Using wska in another project.
Releases are cut with task release; see Releasing.
Install is idempotent — run it twice and the second run changes nothing — and it never writes over a path it does not own. A real directory, or a symlink pointing outside this repo, is refused and named rather than replaced.
Installers run by default from the Task wrapper; pass RUN_INSTALLERS=0 to link
only. The bare CLI is the other way round: installers are skipped unless you
pass --run-installers.
Declared in profiles.toml. Scope is never recorded there, so
one profile serves both --scope project and --scope user.
| Profile | Includes | Adds |
|---|---|---|
base | — | wska, graphify, rtk, ponytail, ponytail-plugin, grill-me, grilling, weebo-testing, weebo-good-practice, nr-check, weebo-hud, weebo-dum-e |
weebo | base | weebo-taskfile-mise, weebo-security, weebo-deps, weebo-containerfile, weebo-api-contract, weebo-release, weebo-cde |
planning | base | bmad, batlehub-rfc, judge |
design | base | frontend-design, theme-factory, impeccable, batlehub-design |
frontend | design | webapp-testing |
mcp | base | mcp-builder, mlab |
game-design | frontend | bmad-gds, blender-image-to-3d, 3dviz-pro-max |
review | base | judge, doc-review, diff-review |
docs | base | weebo-techdoc |
game-design carries BMGD (bmad's game dev studio module) as its own item,
bmad-gds, because the bmad installer takes --modules as the whole install:
that command names bmm,gds so BMM survives. Installing planning and
game-design together runs that installer twice — pick one.
Its two 3D voices are deliberate, the way design carries three UI ones:
blender-image-to-3d is a gated reference-image-to-game-asset pipeline,
3dviz-pro-max is broader scene and Three.js work. The latter is 12 MB of
catalog and kit templates — drop it from the profile if your project has no use
for it.
base carries the four things every project gets: the graph (graphify), the
output filter (rtk), lazy-dev mode (ponytail) and the grilling interview.
grill-me is a three-line shim whose body is "call the Skill tool with
grilling", so the pair travels together. Ponytail's five companion skills are
vendored but not in base — add the one you want to a profile.
Upstream ships ponytail as a Claude plugin whose hooks arm ponytail mode at
session start. Vendoring carries the skills only: ponytail is invoked as a
skill rather than armed by a hook, and ponytail-gain has no benchmark data to
read. base therefore also installs ponytail-plugin, the upstream plugin,
which keeps the hook (and the benchmarks) in project scope.
design overlaps deliberately: frontend-design and theme-factory are
references, impeccable is opinionated. Loading all three gives three voices
that can disagree — trim to the one you trust per project.
PROFILE=all is reserved and means every item in the repo; a profile may not be
named all.
This is the beginning of the README. Read the rest at the original. Read the full README ↗