How This Site Writes Its Own Changelog

Most portfolio sites claim their owner "builds AI agents." This one proves it structurally: the build log on this site is written by an agent, on a schedule, from real commit data, and the whole pipeline is public.

The problem with portfolio claims

A portfolio is a pile of assertions. "I ship weekly." "I automate operations." A visitor can't verify any of it, so the claims compress to noise. The fix is to make the site itself an artifact of the systems it describes.

The architecture

The pipeline is three components, all visible in the repo:

1. A GitHub Action on a Monday-morning cron
2. A Node script that fetches the week's commits across my GitHub account
3. A Claude API call that turns those commits into a short, factual changelog entry

The output is a JSON file committed straight back to the repository. That commit triggers a Vercel deploy, so the /log page, which is fully static, updates itself with zero runtime infrastructure. There is no database and no cron server. Git is the database.

cron (Mon 07:00 UTC)
  └─ fetch last 7 days of commits (GitHub REST)
      └─ Claude: summarize, strict no-invention prompt
          └─ commit src/content/log/2026-W29.json
              └─ push → Vercel builds → /log updates

The part that matters: constraining the model

The interesting engineering is the prompt contract rather than the API call. The agent receives only the commit list and is instructed to summarise nothing beyond it: no inferred features and no marketing language. If a week has three boring dependency bumps, the log says so. An agent whose output you can't trust is a liability wearing a novelty hat.

The same principle runs the rest of this site: the hero terminal replays real commits from the GitHub API, and the status banner genuinely pings my production sites and reports real response times, failures included.

Why this is the pitch

This pipeline is a miniature of what I build for clients: scheduled agents, strict output contracts, no-surprise failure modes, and infrastructure that costs pennies. If you want this pattern applied to your own reporting, digests, or monitoring, that's exactly what the agentic workflow consulting engagement covers.