Claude Code Power Pack
12 tested skills + 4 hooks for Claude Code
I ran an experiment: I picked the problem and made every call on what should ship, then told Claude Code to research a small, genuinely useful thing it could build, build it, and actually test it — not just write code and call it done.
It landed on a skills + hooks pack for Claude Code. Before trusting any of it, I made it prove the pack actually works, not just that it looks right:
It pulled the current docs for the skills system before writing anything — custom commands were recently folded into skills, and a lot of existing content still teaches the old format.
For every skill, it built synthetic git repos and ran the embedded logic against them instead of eyeballing the syntax. That caught a real bug: one skill's base-branch fallback silently produced garbage in a repo with no origin remote and under 10 commits, because git rev-parse HEAD~10 prints its unresolved argument to stdout even when it fails.
For the hooks, it pipe-tested each one with the exact JSON Claude Code sends, down to building a fake Windows .cmd formatter to confirm the auto-format hook actually invokes it — which caught a genuine Windows-only bug (shutil.which() resolves "black" to black.CMD, but re-invoking the unresolved name silently fails on Windows since CreateProcess doesn't do PATHEXT search the way shutil.which() does).
Both are documented with the real before/after in HOOKS.md, included in the download.
What's in it: 12 skills — PR descriptions, pre-commit secrets scanning, test-gap detection, changelog drafting, flaky-test triage, a release checklist, and a few others — plus 4 hooks (auto-format, a bash safety check for force-pushes/hard resets, a secret-write guard, a session-end reminder). One install command, no config file, no dependencies beyond git and Python.
$9 on Gumroad: https://toyanov8.gumroad.com/l/qqzwq
Curious if others have hit that same rev-parse stdout leak, or found other current-vs-outdated-docs mismatches now that skills replaced commands.
3 Comments
The testing layer is more interesting than the 12 skills themselves.
Curious whether buyers are paying mainly to save the implementation time, or because the fact that you actually tested the pack removes the uncertainty of trusting community-made skills and hooks.
Honestly, mostly the second one. Writing 12 skill files isn't hard. Anyone could knock that out in an afternoon with Claude Code itself. What's actually hard to do yourself is the boring part: setting up synthetic repos, hitting edge cases like no origin remote, checking what happens on Windows vs Mac. That's the stuff that takes forever and that most people just skip and hope for the best. So yeah, I think what people are paying for is mostly "someone already found the bugs so I don't have to."