decodingai-magazine/building-a-coding-agent-from-scratch-course/demo-3-repo-pulse

Build a repo pulse dashboard from live data: probe the GitHub API, pull a repository's last 52 weeks of activity, analyse it, and render everything into one self-contained .decode/outputs/dashboard.html the human can open in a browser. All files this demo produces live under .decode/outputs/ (unless the human named a different path).

Default target repository: pydantic/pydantic-ai. If the human named a different owner/repo, use that instead.

1. Probe the data source

Use web_fetch to look at the live API and learn the exact field names before writing any code:

• https://api.github.com/repos/pydantic/pydantic-ai — note stargazers_count, forks_count,

open_issues_count, description.

That one probe is enough — do not fetch the big endpoints with web_fetch; the analysis script will pull those.

2. Write the analysis script

Write .decode/outputs/pulse.py and run it with the one dependency brought in just for the run (do NOT add project dependencies):

uv run --with requests python .decode/outputs/pulse.py

The script must:

1. GET these three endpoints (plain requests, no auth token needed):

• https://api.github.com/repos/{owner}/{repo} — the headline stats.
• https://api.github.com/repos/{owner}/{repo}/stats/commit_activity — the last 52 weeks

of commit counts, one entry per week (total commits + a Unix week timestamp). The whole year arrives in this ONE request — never page through /commits for it.

• https://api.github.com/repos/{owner}/{repo}/contributors?per_page=10 — top contributors

with their commit counts.

1. The stats endpoint answers 202 while GitHub computes the data: retry a few times with a

short sleep until it answers 200 with the 52-entry list.

1. Aggregate: commits per week across the 52 weeks (label each week with the ISO date of its

week timestamp), the busiest week of the year, and the top-10 contributor leaderboard.

1. If the API answers 403 (rate limit), say so plainly; if GITHUB_TOKEN is set in the

environment, send it as a Bearer header.

3. Render the dashboard — charts as inline SVG

Have the script generate .decode/outputs/dashboard.html — one file, zero external requests, no chart library. Draw both charts as inline <svg> elements built with Python string formatting (no image files, no JavaScript):

• A header with the repo name and description.
• Four stat tiles: ⭐ stars, 🍴 forks, 🐛 open issues, 👥 contributors shown.
• Commits per week — an SVG bar chart: one <rect> per week (52 bars), heights scaled to the

busiest week, a handful of month labels along the x-axis, and a <title> tooltip per bar (YYYY-MM-DD — N commits).

• Top contributors — an SVG horizontal bar chart: one <rect> per contributor, with the

login and commit count labelled as <text>.

• Simple clean CSS, dark background.

4. Verify and report

1. Sanity-check the HTML exists and carries both charts:

grep -c "<svg" .decode/outputs/dashboard.html should print 2.

1. Tell the human to open it: open .decode/outputs/dashboard.html.

Report the headline numbers (stars, forks, open issues), the busiest week of the year, the top contributor, and the one-line command to open the dashboard.