"People assume that configuration is a strict progression of cause to effect, but actually, from a non-linear, non-subjective viewpoint, it's more like a big ball of wibbly-wobbly... skilly-willy... stuff."
— a Time Lord, probably, if Time Lords shipped install scripts
There is a particular flavor of pain known only to people who run Claude Code seriously: the slow, soul-eroding ritual of configuring fifty-one little things by hand. You wire one hook. You forget the second. You discover the third only exists in a Discord message from four months ago. Somewhere, a yak grows another inch of fur specifically so you can shave it.
Claude-ITect-Skill
exists to make that ritual unnecessary. It is a one-command starter pack that drops a curated arsenal of skills, agents, and session hooks into any project's .claude/
directory, and then, with the smug confidence of someone who has clearly been burned before, tells you to run /audit
to make sure it actually worked. The author calls people like himself "Claude-ITects™," which the industry refuses to call us, and honestly, after using this, the industry should reconsider.
This is the update, the regeneration, if you will. Same face-of-the-project, new internals. Let's open the TARDIS doors and see how much bigger it is on the inside.
Install it. It deposits a .claude/
folder containing 54 skills, 4 agent definitions, and 6 hook files into your project, patches your settings.json
to wire up the session hooks, and gets out of your way. The README's entire onboarding flow is two words long: run audit
. That restraint is the first sign you're dealing with someone who has actually used the thing he built rather than someone who just wanted a README with a lot of headers.
The install story is genuinely good. PowerShell for Windows, bash for macOS/Linux, sensible --force
and --skip-hooks
flags, and a thoughtful -ProjectPath
option so you can aim it at a project other than your current directory. The -ExecutionPolicy Bypass
note even reassures nervous Windows users that nothing permanent is happening to their system. It's the kind of small kindness that separates a tool someone made for themselves from a tool someone made for other people too.
The 54 skills are organized into sensible tribes, which is where Captain Caveman would put down his club and nod approvingly. (Captain Caaaaaveman! Sorry. It happens.)
Superpowers — the workflow orchestration layer. This is the cleverest part of the whole pack and the part that most people will underestimate. Fourteen skills that auto-trigger at key moments rather than waiting to be summoned. brainstorming
fires before implementation. systematic-debugging
runs a real hypothesis loop. verification-before-completion
is the digital equivalent of a Weeping Angel standing behind you whispering "don't declare it done yet." These aren't tools you reach for; they're behavioral guardrails that shape the agent before it does anything dumb. Borrowed and adapted from the excellent superpowers
project, they're the connective tissue of the pack.
Engineering. Ten skills covering the unglamorous, load-bearing parts of real work: diagnose
(reproduce → minimise → hypothesise → fix → regression-test, a loop with actual discipline), to-prd
and to-issues
for turning a conversation into shippable GitHub work, improve-codebase-architecture
, zoom-out
for when you've been heads-down so long you've forgotten whether the approach was ever correct. This is the section that earns the "architect" in Claude-ITect.
Caveman — token compression. Here is the headline act, and the source of every pun in this review. The caveman family cuts token usage by roughly 75% while keeping full technical accuracy, communicating in a deliberately compressed register with modes named lite
, full
, ultra
, and the gloriously over-the-top wenyan-*
. There's caveman-commit
for compressed commit messages, `caveman-review`
for terse PR comments, and `caveman-stats`
to show you the real token savings from your session log. Unga bunga, big save. This is the feature that justifies the whole bundle existing, because token efficiency isn't a nice-to-have at scale, it's the difference between a session that finishes and a session that runs out of context two steps from the finish line.
The CaveCrew agents. Four subagents, three of them genuinely well-scoped:
file:line
table and, crucially, offers no fix suggestions. It finds; it does not editorialize.The rest: utilities (adr
, audit
, tools
, phase
, `setup-pre-commit`
), productivity (`grill-me`
, the relentless interview skill that stress-tests your plans until they cry), and a small writing suite (edit-article
, writing-beats
, writing-shape
). Plus thefuck
, which diagnoses your last failed shell command and proposes the correct one — and, refreshingly, never executes destructive corrections silently. It confirms first. Even the joke tool has manners.
Three design decisions stand out as evidence of real taste rather than just enthusiasm.
First, audit-first onboarding. Most skill packs assume installation equals success. This one ships duplicate skills on purpose so the /audit
skill has something to react to, verifying that the pack nests cleanly inside whatever skills your repo already has. That's a tool designed to survive contact with a messy real-world project, not a pristine demo repo.
Second, agents that refuse. The Builder's hard cap on file count and the Reviewer's no-praise rule are constraints, and constraints are where quality lives. A sonic screwdriver that did literally everything would be a worse plot device; the good ones have rules.
Third, readable-code discipline runs underneath everything. The karpathy
skill operates as an internal reasoning layer, and the broader architecture leans on three-law readable-code rules and ADR-backed decisions. The pack doesn't just generate code — it generates code it can stand behind later.
The best part about reviewing something built in public is that the rough edges aren't flaws, they're the next update, just sitting there waiting to be claimed. Here are a few friendly ones, offered in the spirit of "Because I ran out of time, and couldn't shipt them myself."
Lock the headline number down. The skill count appears in a couple of places and they don't all agree yet, the live README says 54. For a project whose whole charm is "you don't have to track 54 things by hand," making that number identical everywhere is a five-minute win that'll make the front page feel airtight. Pick the true count, paste it everywhere, never blink.
The em-dash easter egg. v1 shipped a check-encoding.js
hook that cheerfully blocked any file write containing em-dashes or smart quotes. If you want the cosmic balance restored, it's a perfect first task to hand to /audit
. (Or not. Some of us think the dashes earned their place.)
Let the personal bits raise their hand. The Geometry Solver agent and the Project Specific pieces are deeply specific to the author's own engine, and the README is refreshingly upfront that the NgonENGINE commands stay opt-in. That honesty is exactly right. The natural next step is letting the install flow ask which flavor of Claude-ITect you are, so the bespoke bits show up only for the people who'll cheer to see them. The author already wants this thing custom to whoever's using it, which is the whole reason it has a soul in the first place.
None of these are knocks. Every great TARDIS starts as a battered blue box that somebody loved enough to keep flying.
Claude-ITect-Skill
is what happens when someone who actually does the work gets tired of doing the setup for the work; and then, instead of grumbling about it forever, fixes it once, properly, and hands the fix to everyone else for free. That generosity is the throughline. The caveman compression layer alone is worth the install. The superpowers orchestration is the quiet genius humming underneath. The CaveCrew agents demonstrate the rarest virtue in any tool, knowing what to refuse. And the whole pack carries a readable-code, ADR-backed discipline that means it generates work it can still stand behind a month later.
It's an arsenal assembled by one practitioner with real taste and then opened up to the rest of us, inside jokes and all. If you run Claude Code and you've ever felt that yak-shaving despair, that slow drip of one more thing to configure; clone it, run the install, and (say it with me) run /audit
. You'll be set up in the time it takes to make coffee.
Then go build something wonderful. The clubs are optional. The token savings are not. Captain Caaaaaveman would be proud.
Written against the live repository: 54 skills, 4 agents, 6 hooks. Numbers subject to regeneration without notice.
Are you a Claude Architect or as the industry refuses to call us: Claude-ITects™ Good. You're in the right place.
Welcome to Claude-ITect-Skill The "I don't want to configure 54 things manually" starter pack for Claude Code. Powered by Captain Caveman energy -- one club, infinite tools, zero yak-shaving.
What you get (whether you deserve it or not): But don't get distracted by the features.
The real sorcery happens during setup when everything just… works.
No funny dances. No ritual sacrifices.
Install it. Try it. Judge me harshly afterward.
After installation start with these skills:
/onboard
<- run first/audit
<- run next