connor

Ask first

• What kind of league? (default a fantasy league; an office pool, golf league or rec league with weekly scores works the same)
• Where do the scores live? (default your league site; a shared spreadsheet works too)
• Is your league public, or does it need a login to read? (default public)
• How should teams be ranked? (default ranking points: each week every team earns points by where its score placed that week, summed over the season)
• How many teams make the playoffs? (default 6)
• How often should it refresh? (default every 2 hours on game days, once a day otherwise)
• Where should the page live? (default a free static site host)

Needs

• your league's weekly scores, from the league site's public data or a page you can read.
• a free place to host a static page and run a scheduled job (a code host with scheduled jobs works).
• If your league is private, a login the job can use; keep it in the host's secret store, never in the code. It reads scores and team names only.
• If your assistant can only chat: paste the weekly scores each week and ask it to return the updated standings table.

Build

1. Read: on schedule, fetch every completed week's scores for every team.
2. Decide: for each week, rank teams by points scored (highest = N-1 points, lowest = 0; ties split the points). Sum per team across weeks.
3. Decide: find the team sitting at the last playoff spot and subtract its total from everyone, so the cutoff team shows 0, teams above are positive, teams below negative.
4. Decide: only count weeks marked final; skip the current week until it ends. If nothing changed since the last run, stop without committing.
5. Deliver: write the table to a data file, rebuild the page, and publish it. Show "last updated" with the date and time.

Check

• Run it once by hand on last season's data; the cutoff team must read exactly 0 and the totals must add to a known week-by-week count.
• A quiet run in the off-season changes nothing and publishes nothing.
• Confirm the first scheduled run by checking that the "last updated" time moves after the next refresh.

In my version

• ESPN league, a static site on GitHub Pages, refreshed by a GitHub Actions job called "Update League Data".
• I am not an engineer; Claude Code wrote all of it from my description.

Ask first

• Which fantasy platform and scoring? (default head-to-head categories; the same plan works for a weekly move limit in any sport)
• Is there a weekly cap on pitcher starts? (default 8 starts a week)
• What makes a streamer worth it? (default ERA under 4.20 and WHIP under 1.30 over the last 30 days, facing an offense in the bottom half by runs scored)
• How many streamers to show? (default top 10)
• When should it refresh? (default once a day, early morning, during the season)

Needs

• probable starting pitchers for the next 7 days, from a public baseball data source.
• recent pitcher stats and team offense rankings, from a public stats source.
• who is a free agent in your league; if the platform cannot be read, paste your roster and rostered players once a week.
• If your assistant can only chat: each Monday paste the week's probable starters and your roster, and ask for a start plan that stays under your cap plus the top 5 free agents to stream.

Build

1. Read: on schedule, pull probable starters for the next 7 days, each pitcher's last-30-day stats, and each opponent's runs per game.
2. Decide: count starts already used this week (Monday to Sunday) against the cap. Two-start pitchers count twice.
3. Decide: score each free-agent starter: pass only if both stat thresholds hold; rank passers by opponent weakness, then by number of starts left this week. Drop anyone whose start is already in the past.
4. Decide: for your own pitchers, flag starts to bench if they would push you over the cap; keep the higher-ranked start.
5. Deliver: update the planner page with starts used / cap, your recommended starts, and the streamer list. The person makes every roster move themselves.

Check

• Run it against last week's data with today set to a past Monday; starts counted must match the box scores, and no listed streamer may be rostered.
• A day with no new probables changes nothing.

In my version

• A paid web app I made with Claude Code (Next.js and a database), $5 a month, refreshed by a daily scheduled job.
• 8-start weekly cap, built for head-to-head category leagues.

Ask first

• What should the app do, in one sentence? (default a page that shows one number you check often)
• Who uses it? (default just you)
• Should it update on its own? (default yes, once a day)
• Where should it live? (default a free static host)

Needs

• a coding agent that can write files and run commands on your computer.
• a free code host if it should be online.
• If your assistant can only chat: ask it for a single HTML file you can open in a browser, and paste errors back to it.

Build

1. Describe the outcome, not the code: what you see, where the data comes from, how often it changes.
2. Have the agent write a short plan and a backlog file before it writes code; approve the plan.
3. Have it build the smallest version that shows real data, then run it and show you.
4. Add a scheduled job only once the manual run is right.
5. Keep secrets in the host's secret store; the agent never writes them into code.

Check

• You can open the app and see real data, and the agent can explain each file in one line.

In my version

• Claude Code for everything. I am not an engineer; the league site, a paid fantasy baseball app and a seating-chart tool all came out of this.

Ask first

• Who will use it? (default a wedding planner or a couple planning their own wedding)
• How many guests and tables, at most? (default 200 guests, round tables of 8 to 12)
• Which views matter? (default a floor plan and a list by table)
• Where should the guest list come from? (default a pasted list or a spreadsheet upload)
• Should it save online or in the browser? (default in the browser, so no accounts are needed)

Needs

• a guest list: name, party or group, and optional notes like dietary needs.
• a coding agent to build the tool, and a free static host if others should use it.
• If your assistant can only chat: paste the guest list and table sizes, ask for a table-by-table list, and keep a running list of who is not seated yet.

Build

1. Import: read the guest list from a paste or spreadsheet; group guests by party so couples and families move together.
2. Lay out: draw tables on a floor plan; each table can be resized to any seat count, and the seats update to match.
3. Seat: drag a guest or a whole party onto a table. Block a drop that would put more people at a table than it has seats, and say why.
4. Track: keep a live "not seated yet" list and a count of open seats per table, so nothing is missed.
5. Views: switch between the floor plan, a list by table, and a list by guest; let the user filter by group.
6. Save and export: save as they go, and export a printable list by table for the venue.

Check

• Load a made-up list of 60 guests, seat 50, and confirm the "not seated yet" list shows exactly the other 10.
• Shrink a full table by two seats; the tool must flag the two extra guests, not drop them silently.

In my version

• A web tool I built with Claude Code for wedding planners, pre-launch, with a public demo using made-up guests. Edits save in the browser.

Ask first

• Scoring format? (default half PPR)
• League size and roster spots? (default 12 teams, standard roster)
• Which betting lines to use? (default season-long player props: rushing, receiving and passing yards and touchdowns)
• Should it blend with expert rankings? (default no, odds only)

Needs

• current season-long player prop lines from a public odds source, or copied from a sportsbook page.
• If your assistant can only chat: paste the prop lines for the top 200 players and your scoring rules, and ask it to convert each line to projected points and rank them.

Build

1. Read: collect every available season-long prop for each player, using the middle of the over and under.
2. Decide: convert props to fantasy points with your scoring (for example 0.1 per yard, 6 per touchdown, 0.5 per catch). Players missing a prop get a positional average for that stat, flagged.
3. Decide: rank within position, then overall by points above the replacement player at each position (the player ranked just past the last starter in a league your size).
4. Deliver: a sortable table or spreadsheet with overall rank, position rank, projected points and which numbers were filled in.

Check

• Spot-check five well-known players by hand; projected points must match the arithmetic from their lines.
• Every filled-in number must be flagged in the output.

In my version

• A Python project Claude Code built for me, run by hand in the weeks before my draft. Half PPR, 12 teams.