When Claude’s Bad Schedule Made Me Laugh, I Built My First No‑Code Skill

What happened

Yesterday I asked Claude Code to schedule a day – big cleaning, a park walk, learning tech, and a deployment. Claude earnestly replied with a plan:

8 am wake up, start cleaning, park in the evening.

Looking at the output I realized several problems.

Why the plan was wrong

1. I actually wake at 10 am. Claude assumed an 8 am start.

2. Beijing evening 31 °C with 40 km/h wind. A park walk is more survival than leisure.

3. “Big cleaning” is only four characters. If I could clean the whole house in an afternoon, why hire a cleaner?

4. No meals were scheduled. Claude apparently thinks I don’t need to eat.

I corrected each line and stopped at the third item, realizing the output was essentially an SOP.

The root cause is that a one‑sentence prompt leaves Claude to guess, just like telling an intern “do this project”.

SOP can become a Skill

A Skill is a packaged workflow that can be invoked with /skill-name. I thought: if I standardize “plan a day” so Claude always checks the weather, asks my wake‑up time, and breaks tasks, I won’t have to be a “human error‑corrector” each time. I named the skill smart-planner.

Process: crash → review → copy → validate

Step 1 – Review why it crashed

I identified five “first‑principles” for planning:

Data first – Claude won’t fetch weather unless told.

Human‑centered – Claude assumes a 6 am superhuman schedule.

Constraint‑driven ordering – Schedule immutable items (weather windows, fixed appointments) before filling the rest.

Task granularity – “Big cleaning” must be split into living‑room, kitchen, bathroom, bedroom.

Leave buffer – Never fill 100%; tasks often take 1.5× the estimated time.

Step 2 – Design the workflow

Based on the principles I built a nine‑stage Q&A flow:

Stages 0‑3 (information collection) : diagnose current state, collect tasks, lock priorities, interview.

Stages 4‑6 (data‑driven) : actively fetch weather/traffic/server status, extract hard constraints, break tasks.

Stages 7‑9 (output loop) : allocate time, output plan, evening review.

The key is step 4: when the user says “go to the park”, Claude runs curl wttr.in to fetch the weather automatically, without me having to prompt for it.

Step 3 – Copy existing Skills

I searched the market with npx skills find and found four useful skills:

From plan‑my‑day – the Must/Should/Nice priority levels.

From daily‑rhythm – the evening review template.

From personal‑productivity – diagnostic questioning (used by 4,300+ installs).

From goal‑planner – the Goal vs Task distinction.

“Building a product is not from 0 to 1, but from 0 to ‘others’ good stuff + my improvements’.”

Step 4 – Validate

I reran the same scenario:

“Tomorrow: big cleaning, park walk, learn tech, deployment.”

Before (bare version) : Claude output “8 am wake up, cleaning first, park in the evening”.

After (with Skill) : Claude first fetched Beijing’s weather (cloudy 25‑28 °C, comfortable morning), asked my wake‑up time, park duration, and even inspected my GitHub repo to find the “pet‑comfort‑card” project. The final schedule:

10:45 am – Park ("Morning 25 °C breeze, best outdoor window").

1:00 pm – Cleaning ("Noon 28 °C wind, indoor work unaffected").

4:00 pm – Learning ("Second energy peak, suitable for deep reading").

7:15 pm – Deployment ("Quiet environment, focus on hands‑on work").

Each block includes an explanation derived from data, not a guess.

Three visible changes

I no longer act as a “human error‑corrector”. The skill asks the right questions and fetches data up front, saving me enough time to watch three short videos.

The plan respects human constraints – weather, meals, fatigue – instead of a naïve programmer’s schedule.

The Skill is a single file on GitHub; I can iterate it daily. Unlike a chat log, the code lives on github.com/yinshaojun001/smart-planner.

Final thoughts

AI’s upper bound is high; its lower bound depends on the constraints you give it.

If you tell AI “help me plan”, you get a blob. If you tell it “plan, first check weather, then ask my routine, then allocate according to energy curves”, you get a usable schedule. The difference lies in solidifying good prompting as enforceable rules inside a Skill.

Install the skill with a single command:

npx skills add yinshaojun001/smart-planner -g -y