· LAB
I saw this in May:
CJ Zafir @cjzafir It was a simple /goal to make my internet network faster.
> Diagnosed the issue: ran speedtest-cli first.
> Checked DNS resolution times,
> Checked MTU, packet loss, Wi-Fi signal/interference.
> Found 3 issues.
> Removed stale network locations/profiles.
> Killed or throttled bandwidth-hogging background processes.
> Optimized mDNS.
> Ran before/after speed tests and latency checks.
So I dropped the link into my Claude Code and asked it to do the same for me.
I pay for a gigabit. My laptop at the desk showed 100 Mbps. For two years I assumed that’s just what Wi-Fi is.
The skill
Those nine days turned into check-network. Measure before recommending: throughput, how much latency jumps under load, packet loss, what rate the Wi-Fi link actually negotiated. Then localize across four layers — ISP plan, router, Wi-Fi path, client. Then a ladder of fixes where a cable is always first and a second wireless extender is never.
Install
Claude Code:
/plugin marketplace add yermilov/learn-yy-skills
/plugin install home-it@learn-yy-skills
In Codex there’s no command to look up — launch it and just ask: “install the marketplace yermilov/learn-yy-skills and the home-it plugin from it”. It takes it from there.
Or take the single skill into any agent, no marketplace (npx skills supports Claude Code, Codex, Cursor and 60+ others):
npx skills add yermilov/learn-yy-skills --skill check-network
Then just say “make my home network faster”. Works on macOS, Windows and Linux — different tools, same four measurements.
What it found
The first number is a range, and that is not a typo. At the desk the speed wandered between 75 and 190 Mbps from run to run.
The agent ruled out everything I suspected. The plan is already the top tier. DNS answers in 10 ms. MTU is fine, zero interface errors. But the ping to my own router was losing 5% of packets, and that ended the argument. If packets die inside the apartment, the ISP has nothing to do with it.
The cause was the extender. The TP-Link I once added “for better coverage” was itself sitting on Wi-Fi: one hand holding the link to the router, the other to the laptop, and each hop halves the bandwidth. The link rate at my desk negotiated 234 Mbps — about what the speed tests were showing. The Wi-Fi had already told me everything, I just never looked.
The cheapest diagnostic in the world: pick up the laptop and walk five meters to the router. −20 dBm instead of −68, an 866 Mbps link, 624 Mbps, zero loss.
The two things the internet tells you to do first did nothing. Turning off Smart Connect: 611 vs 612 Mbps. Picking “cleaner” channels by hand: 398 and 418, against 624 on auto — worse.
I asked an LLM council (ChatGPT + Gemini) whether to buy a second extender for the desk. Both ranked that option last: another wireless hop halves the bandwidth again. The right fix is a cable from the router to the desk with the extender in access-point mode.
I never ran that cable. I plugged the Mac into the extender’s gigabit port — the same extender still hanging off a wireless backhaul — so roughly 80% of the fix. 492 Mbps, zero loss. The agent honestly noted in its report that this is not a real wired backhaul and that 800–940 Mbps is still on the table.
500 Mbps is quite good already. I stopped there.