尚未偵測到更新。GitHub 發布版本後會顯示在這裡。
個人 GitHub 帳號下公開儲存庫的擁有者可以登入並驗證所有權,自行管理介紹,無需管理員核准。
登入並產生驗證檔案,在 30 分鐘內將其提交至此儲存庫的預設分支。其他分支或分叉儲存庫中的檔案無效。然後在下方檢查檔案,即可無需管理員核准直接管理專案。請保留檔案以維持管理權限。
如需登入,將前往登入頁面。登入後請回到此處,再次點擊所選的驗證按鈕。
在專案資料夾中執行 /nomorevibe verify,即可驗證所有權並自行管理介紹。
一分鐘完成驗證A Fantasy Premier League agent that recommends personalized transfer and captain picks, as well as chip strategies.
翻譯準備中
取自儲存庫 README
正在載入完整 README。 以下為已儲存的摘錄。
An agentic Fantasy Premier League advisor with a Premier League pundit's eye — grounded in real data, not vibes.
Every gameweek, millions of FPL managers face the same dread: a transfer to make (or two, or five), a captain to pick, a chip you're terrified to waste — and a deadline counting down. So you crowd-source it: group chats, Reddit, ten browser tabs of conflicting "templates," none of which know your squad, your bank, or how many free transfers you're actually holding. You make the call on a hunch and find out on Saturday.檢視完整 README ↗An agentic Fantasy Premier League advisor with a Premier League pundit's eye — grounded in real data, not vibes.
Every gameweek, millions of FPL managers face the same dread: a transfer to make (or two, or five), a captain to pick, a chip you're terrified to waste — and a deadline counting down. So you crowd-source it: group chats, Reddit, ten browser tabs of conflicting "templates," none of which know your squad, your bank, or how many free transfers you're actually holding. You make the call on a hunch and find out on Saturday.
Pocket Scout is your personal scout. Enter your FPL manager ID and it reads your team and tells you the highest-leverage move this week — transfers, captaincy, chips — and explains why, the way a post-match analyst breaks down a game. It's personalized (your squad, your bank, the exact free transfers you hold), educated (grounded in expert FPL principles and the live rules), and deterministic (a reproducible engine does the math — same squad, same answer — not an LLM guessing).
https://github.com/user-attachments/assets/93f5c5eb-443b-4edb-b4d6-2a8c6037cfe0
One clear answer at the top. The first thing you see is this week's advice in a single line: who to sell and who to buy, who to captain, and whether to play a chip. A button beside it takes you straight to the FPL transfers page.
A rating for every player in your squad. Each player on the pitch gets a score out of 10. It starts from FPL's own points prediction for the next match and adjusts it for recent form, upcoming fixtures, price, and how well the player is really performing underneath the headline numbers.
Ask The Scout. This is the heart of the app. The Scout greets you with a short brief for the coming deadline, then you can ask it anything in plain words: "Should I swap Wissa for Emersonn?", "Who should I captain?", "Is it worth taking a hit?" It looks up the real numbers before it answers, so it never makes things up, and what it says always matches the advice on screen.
Act on it. Click any player, on the pitch or in a suggestion, for a quick card: age, nationality, recent form, last week's minutes and points, and expected points next week, with a link to their Premier League profile. When you are ready, the banner at the top hands you over to FPL to make the move.
The full breakdown, in a drawer you open when you want the detail, in three tabs:
Try it without a team. No FPL ID? Pocket Scout builds a sample squad from the best players in the game, and you can still ask the Scout anything about FPL. In this mode there is no personal transfer plan, just the chat, and it always answers from this season's rules.
Everything comes in one voice, Pocket Scout, backed by expert FPL knowledge: when to play chips, how player ownership affects your rank, and the official rules.
%%{init: {"flowchart": {"curve": "monotoneY", "nodeSpacing": 60, "rankSpacing": 80, "subGraphTitleMargin": {"top": 12, "bottom": 12}}}}%%
flowchart TB
subgraph IN["Your input"]
YOU(["Your FPL manager ID"]):::you
DEMO(["No ID?<br/>Explore a sample squad"]):::you
end
FPL[("Official FPL data")]:::data
NEWS[("Injury and team news")]:::data
subgraph P1["Phase 1 — Rank Your Squad"]
RANK["Rates every player 0–10<br/>pure math"]:::math
end
subgraph P2["Phase 2 — Compute Transfer and Chip Strategy"]
PLAN["Transfers, captain, chips<br/>pure math"]:::math
AI["Explains the plan<br/>Claude AI"]:::ai
CACHE[("Saved for 10 minutes")]:::cache
PLAN --> AI --> CACHE
end
subgraph SCREEN["Your screen"]
OUT["Verdict · pitch · ratings<br/>breakdown"]:::screen
CHAT["Ask The Scout chat<br/>Claude AI · real numbers"]:::screen
end
YOU --> P1
DEMO --> P1
FPL --> P1
RANK --> P2
NEWS -.-> P2
RANK -->|instantly| OUT
CACHE -->|once ready| OUT
CACHE -.-> CHAT
KB[("FPL expert playbook")]:::data -.-> CHAT
classDef you fill:#4f81bd33,stroke:#4f81bd,stroke-width:1.5px
classDef data fill:#3a9c8f2e,stroke:#3a9c8f,stroke-width:1.5px
classDef cache fill:#3a9c8f2e,stroke:#3a9c8f,stroke-width:1.5px,stroke-dasharray:4 3
classDef math fill:#c9822b2e,stroke:#c9822b,stroke-width:1.5px
classDef ai fill:#8b6cc42e,stroke:#8b6cc4,stroke-width:1.5px
classDef screen fill:#5aa05a2e,stroke:#5aa05a,stroke-width:1.5px
style IN fill:#4f81bd14,stroke:#4f81bd80
style P1 fill:#88888814,stroke:#88888880
style P2 fill:#88888814,stroke:#88888880
style SCREEN fill:#5aa05a14,stroke:#5aa05a80
Color key: blue = your input · teal = data · amber = pure math · purple = Claude AI · green = your screen.
No manager ID? Phase 1 rates a sample squad instead, Phase 2 only picks a captain (no transfer, long-term or chip plan), and Ask The Scout answers general FPL questions grounded in this season's rules.
Phase 1 is pure math, so the pitch paints immediately. Phase 2 does the heavier planning, then Claude writes the reasoning (the chip plan is grounded in the same expert playbook the Scout uses), and those insights stream in while you read; the Scout opens with a brief for the coming deadline. → Full breakdown: docs/ARCHITECTURE.md
Before the app gives advice, that advice is tested against ten seasons of real FPL history to see whether it would actually have earned points. Decisions about the app are made on that evidence, including when the evidence says no.
→ The full story (backtests, replays, the no-ship): docs/EVALUATION.md
| Tell it your situation — enter 0–5 free transfers | This Week — Make N free transfers + the ep-native Restructure |
|---|---|
Expert, grounded chat — no hallucinations, no prompt injection. Ask The Scout reasons over real numbers it fetches via tool calls, stays anchored to your committed plan and the curated knowledge base, and refuses to be talked out of the facts — so its answers never drift from the panels or invent stats.
| Ask The Scout — a tool-grounded answer citing real numbers | …and it holds the line against off-topic / injection attempts |
|---|---|
| Player detail dialog — click any player to inspect | Get scouted — enter your manager ID, or explore without one |
|---|---|
| Explore without a team — the demo dream team + Scout chat |
|---|
You need an Anthropic API key for the AI features. It also runs without one — the pitch, ratings, and deterministic recommendations work; only the LLM prose falls back.
Option A — dev (simplest):
git clone https://github.com/LionelKavit/Fantasy-Premier-League-Advisor.git
cd Fantasy-Premier-League-Advisor
npm install
echo "ANTHROPIC_API_KEY=sk-ant-..." > .env.local # optional
npm run dev
# open http://localhost:3000 (enter any public FPL manager ID — or click "Explore without a team")
Option B — production build (Docker + Caddy):
# create .env.docker (gitignored) with at least your key:
# ANTHROPIC_API_KEY=sk-ant-...
# SITE_ADDRESS=:80
# BASIC_AUTH_USER=admin
# BASIC_AUTH_HASH=... # see note below
docker compose up --build
# open http://localhost (log in if you set basic-auth)
Note on the password hash: generate it with
docker run --rm caddy:2-alpine caddy hash-password --plaintext 'yourpass', then double every$to$$in.env.docker(Docker Compose otherwise reads$as a variable). Details in docs/ARCHITECTURE.md.
Next.js 16 (App Router) · React 19 · TypeScript · Tailwind + shadcn · Anthropic Claude (Sonnet, with prompt caching) · Vitest (331 tests) · Docker + Caddy · the public FPL API.
openspec/changes/archive/ (the paper trail behind every feature and eval decision).| Doc | What's in it |
|---|---|
| docs/ARCHITECTURE.md | The decision pipeline, the agentic chat, knowledge grounding, deployment |
| docs/EVALUATION.md | The backtest harness, the data-fit model, squad replays, and the honest no-ship |