sxsphinxstack

Skills / Start a project

Start a project

Pick one project and start it today. Interview, converge on a week-sized idea, create the repo, hand off to the right build skill. Use when they say "I want to build something", "I need something for my resume", or cannot decide between ideas.

Use this skill. Nothing to install.

Open in ChatGPT Open in Claude prefilled and ready; just hit send
for Gemini, Copilot, or Cursor: paste it, then say "use the start a project skill"

Don't have an agent?  ·  Raw file: skills/start-a-project.md

Watch the first two minutes

This is how a session goes when someone arrives with ten ideas and zero started. Click through it.

​(pastes the skill) i want to build something real but i have like ten ideas and i keep switching between them
Ten ideas usually means zero finished, so we converge on one today and start it before we're done. First: what do you actually do with your time? Games, sports, a job, a fandom, a family business?
mostly climbing. i also run the discord for my gym's youth team
That Discord is real material, better than any generic idea. How many hours a week can you honestly give this, and what do you want at the end: a resume line, a portfolio piece, money?
maybe 5 hours. resume line, i'm applying for internships next fall
Two candidates then. A bot for your team Discord that posts session signups, or a site listing every climbing spot in your county, live at a URL. Week one of the bot: it answers one command in your real server. Which one would you be embarrassed to show a friend unfinished?
the bot, honestly. people in the server would ask me about it
The embarrassment means you care, so the bot wins. I'm creating the repo now with a README in your words, plus issue #1: "bot replies to !sessions with this week's schedule." We do the first fifteen minutes together before you go, so the repo isn't empty when you walk away.

Scripted example of a real session.

Too many ideas to pick one?

The skill treats that as the normal starting condition. It interviews you about what you already do with your time, because the project comes from that list. Candidates get one sentence each, with what will exist at the end and what week one looks like. If you stall, it asks which one you would be embarrassed to show a friend unfinished, and that one wins. If you have zero ideas instead of ten, it walks the idea bank with you and bends a shape around your real material. Three shapes that read well later:

Discord bot your server uses Price-drop watcher on a schedule Browser game at a URL

What you end up with

A decision would evaporate by the weekend, so the session ends with a started repo instead. This is the state of things when the conversation above closes:

Example
teambot/
├── README.md          committed
└── bot.js             first 15 minutes done, replies to !ping

README.md
  # teambot
  Discord bot for my climbing gym's youth team server (32 members).
  Posts the week's session schedule and takes signups, so the
  pinned-screenshot system can retire.

Issue #1 — first milestone
  "Bot replies to !sessions with this week's schedule, hardcoded."
  Done when: someone besides me runs the command in the real server.

Handoff: ship-on-github once the bot answers for real, so the
work lives at a public URL. The schedule stays hardcoded until a
real member asks for more.

Questions people actually ask

What if I have no ideas at all?

Then the interview matters more. The skill asks what you do with your time and reads the idea bank at /ideas/ against your answers. The bank is a menu of shapes; the skill bends one around your actual life, your server, your team, your family's business.

Do I have to finish the whole thing in a week?

No. The rule is that week one produces something visible, and the skill cuts features until that is true. A project you can complete outranks a project that impresses in description.

I always abandon projects. Why would this be different?

The skill asks what you already tried and why it died, and treats the reason as a constraint on the next pick. It also refuses to end at a decision: the repo exists and the first fifteen minutes are done before the session closes, which is further than most abandoned projects ever got.

Do I need to know how to code first?

No. Picking and starting need honest answers from you. The building happens in the skill this one hands off to, where the agent writes code and you make the calls.

Is this free?

The skill is free, no account, no catch. The agent you paste it into may or may not be: Gemini has the widest free tier, and the free plans of ChatGPT, Claude, and Copilot can all run it in a chat.

Where to go from here

Project picked? These are the skills it hands off to:

Build a website Build a web app Ship on GitHub

Project not software? These begin where this one ends:

Start a newsletter Start a YouTube channel Start selling something
Curious? Read the full skill — the exact instructions your agent gets 
---
name: start-a-project
category: start
description: Pick one project and start it today. Interview, converge on a week-sized idea, create the repo, hand off to the right build skill. Use when they say "I want to build something", "I need something for my resume", or cannot decide between ideas.
---

# start-a-project

Help someone pick their first real project. They probably
have either too many ideas or none. Your job is to converge on one
project rooted in their actual life, scope it to about a week, and
start it before the conversation ends. Picking is not the deliverable;
starting is.

## Interview

Ask plainly, a few questions at a time:

- What do they actually do with their time? Games, sports, music, a
  job, a fandom, a community, a family business. The project comes
  from this list, never from a generic idea.
- How many hours a week can they honestly give it, and for how long?
- What do they want at the end — a resume line, a portfolio piece,
  money, an audience, a college essay topic?
- What have they already tried and abandoned, and why? The reason it
  stopped is a constraint on the next pick.

Then fetch https://sphinxstack.com/ideas/ and read it against their
answers. The idea bank is a menu, not an assignment: use it to show
them shapes of projects, then bend one around their real material.

## Converge

Put two or three candidates in front of them, each in one sentence
with what will exist at the end ("a site listing every climbing spot
in your county, live at a URL"). For each, say out loud what week one
looks like. Then they choose. If they stall, ask which one they would still
care about after a week of ordinary work, and which visible result they most
want to show.

Scope the winner down until week one produces something visible. Cut
features, not the finish line. A project they can complete outranks a
project that impresses in description.

## Start it now

Do not end at a decision. Before the session closes:

1. Create the repo, named after the project, with a README that
   states in their words what it is and who it is for.
2. Write the first milestone as a GitHub issue: one concrete thing,
   achievable in the first sitting.
3. Do the first fifteen minutes of that milestone together, so the
   repo is not empty when they walk away.

## Hand off

Route them to the skill that builds the thing:

- A site for a person, club, or business —
  https://sphinxstack.com/skills/build-website/
- Something interactive with data or accounts —
  https://sphinxstack.com/skills/build-web-app/
- Getting the work public and deployed —
  https://sphinxstack.com/skills/ship-on-github/

If the project is not software (a newsletter, a channel, a club, a
thing to sell), the start-a-newsletter, start-a-youtube-channel,
start-a-club, and start-selling-something skills each begin where
this one ends. Say which one and why.

## Done

- One project chosen from real interests with a week-sized visible finish line
- Repo, README, and first milestone exist; the first working change is committed
- Time budget, next session, proof artifact, and handoff skill are recorded