NIK Profile Engine

Most profile READMEs stitch together third-party image services. They rate-limit, go down, and look like everyone else's. profile-engine renders every visual itself, inside your own workflow, from live data — and publishes the result to a branch of your profile repository. Nothing on your profile depends on someone else's server at view time.

• Day & night. Every visual is rendered twice and swapped with prefers-color-scheme.
• Self-contained SVGs. Fonts are embedded as base64 and icons are hand-drawn paths, so images never load external resources — exactly what GitHub's image proxy requires.
• Live data. Market prices with 24h sparklines (CoinGecko), Ethereum gas from public JSON-RPC nodes, followers from the GitHub API.
• AI, without secrets. The weekly summary uses GitHub Models with the workflow's own token — no API key to manage — and falls back to a deterministic summary if the model is unavailable.
• Resilient by design. Each module is isolated: one failing source never breaks the others, and the last good snapshot is reused (marked delayed) when an API is down.
• Accessible. Every SVG has role="img" and a descriptive <title>; animations respect prefers-reduced-motion.
• Zero dependencies. Pure Python standard library.

flowchart LR
    T[profile.toml] --> E{{profile-engine}}
    GH[(GitHub API)] --> E
    CG[(CoinGecko)] --> E
    RPC[(Ethereum RPC)] --> E
    GM[(GitHub Models)] --> E
    RSS[(RSS / Atom)] --> E
    E -->|hero · headings · focus · badges · activity · ticker · weekly| S[light + dark SVGs]
    S --> O[output branch]
    O --> R[README via <picture>]

# .github/workflows/profile.yml
name: Profile
on:
  schedule:
    - cron: "7 * * * *"   # hourly: markets ticker
  workflow_dispatch:

permissions:
  contents: write   # publish the SVGs
  models: read      # weekly AI summary via GitHub Models

jobs:
  render:
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v5

      - name: Restore previous output (keeps engine state and caches)
        run: |
          mkdir -p dist
          git fetch --depth=1 origin output && git archive FETCH_HEAD | tar -x -C dist || true

      - uses: LucNIK/profile-engine@v1
        with:
          config: profile.toml
          output_dir: dist

      - uses: crazy-max/ghaction-github-pages@v5
        with:
          target_branch: output
          build_dir: dist
          jekyll: false
        env:
          GITHUB_TOKEN: ${{ secrets.GITHUB_TOKEN }}

Then reference the images in your README, one <picture> per visual:

<picture>
  <source media="(prefers-color-scheme: dark)"  srcset="https://raw.githubusercontent.com/USER/USER/output/ticker-dark.svg">
  <source media="(prefers-color-scheme: light)" srcset="https://raw.githubusercontent.com/USER/USER/output/ticker-light.svg">
  <img alt="Live markets" src="https://raw.githubusercontent.com/USER/USER/output/ticker-light.svg">
</picture>

Visitors click a column button in your README. That opens a pre-filled issue titled connect4|drop|<n>; a workflow plays every open move issue in order (so simultaneous players never lose a move), the engine answers with minimax + alpha-beta pruning (6 plies by default), the board is re-rendered, and the issue gets a reply and is closed.

# .github/workflows/game.yml
name: Connect Four
on:
  issues:
    types: [opened]
permissions:
  contents: write   # publish the new board
  issues: write     # reply to and close move issues
concurrency:
  group: profile    # same group as the profile workflow: one writer at a time
  cancel-in-progress: false
jobs:
  move:
    if: startsWith(github.event.issue.title, 'connect4')
    runs-on: ubuntu-latest
    steps:
      - uses: actions/checkout@v5
      - run: |
          mkdir -p dist
          git fetch --depth=1 origin output && git archive FETCH_HEAD | tar -x -C dist || true
      - uses: LucNIK/profile-engine@v1
        with:
          modules: game
          game_process_issues: "true"   # plays every open move issue, replies and closes it
      - uses: crazy-max/ghaction-github-pages@v5
        with: { target_branch: output, build_dir: dist, jekyll: false }
        env: { GITHUB_TOKEN: "${{ secrets.GITHUB_TOKEN }}" }

Each column button links to https://github.com/USER/USER/issues/new?title=connect4%7Cdrop%7C<n>. Untrusted input is handled defensively: the title must match a strict pattern, player names are validated, and neither ever reaches a shell or the rendered images unchecked.

See examples/profile.toml for a complete file.

[profile]
user = "LucNIK"
name = "Luc NIK"
website = "https://ayuefu.com"
timezone = "Asia/Shanghai"

[theme]
font = "IBM Plex Sans"      # any Google Font
light_accent = "#B7860B"    # day
dark_accent = "#C1121F"     # night

[hero]
lines = ["Luc NIK", "Software Engineer", "AI/ML · FinTech · Web3"]

[headings]
titles = ["About", "Focus", "Markets", "This Week"]

[[focus.items]]
title = "AI / ML"
lines = ["Models · data pipelines", "intelligent applications"]
icon = "network"            # network, candles, ethereum, cloud, gas, spark, globe, users, coin

[activity]
repos = 3
events = 5
rss = ""                    # optional: "https://example.com/rss.xml"

[game]
repo = "LucNIK/LucNIK"      # where move issues are opened
depth = 6                   # engine strength (1-8)

[ticker]
currency = "usd"
gas = true
assets = [{ id = "bitcoin", symbol = "BTC", name = "Bitcoin" }]

[weekly]
model = "openai/gpt-4.1-mini"
exclude_repos = []

python -m profile_engine --config examples/profile.toml --out dist --offline   # sample data, no network
python -m unittest -v

• 1.2.1 — Node 24 runtime (setup-python@v6), commit policy and enforcing git hooks, SECURITY.md.
• 1.2.0 — game module: Connect Four against a minimax AI, played by visitors through issues; GitHub Models call hardened (redirect and non-JSON detection, fallback endpoint); action inputs passed via environment only.
• 1.1.0 — activity module (recent projects, public events, RSS/Atom posts); variable fonts embedded once (≈50% lighter SVGs); detailed GitHub Models errors stored in the engine state, with an automatic retry every 6 hours.
• 1.0.0 — first release: hero, headings, focus, badges, ticker, weekly AI summary.

MIT © John Luke NIKABOU