Goal

After merging PR #10, the author published the two articles by following articles/PUBLISHING.md, and asked Claude to do everything else (translated from Italian). During the publication they asked two quick questions: whether status: ready was right, given that the checklist’s first step says review, and why fig-grid-paths.png isn’t in piece 1.

What we did

  • The piece-01-v1 tag (annotated, on d8687ce, the merge of PR #10). It’s the state in which a fresh Codespace passes everything, make lint included.
  • A check of every link in both articles while the author was publishing. All ten external links returned 200, and all the images exist. The site had already redeployed from d8687ce.
  • The author published piece 1, then the overture, in Docker Captains Wisdom, by 2026-10-31:
  • The repository follows Medium: both articles are published, with their URLs. Their series lists carry the links, example 1 is frozen at the tag, and the README links both articles. The site’s Home page turns medium_url into links by itself.
  • Two badges from the CLAUDE.md §9.3 backlog:
    • “Latest tag” (25500c3). It sorts by date, because the tags aren’t semver.
    • “Articles published 2/7”. shields.io can read a value from a YAML file but can’t count, so series.yml gained a published: "N/7" field, and make site-check fails if it disagrees with the statuses (0a10cb0). Tested with a correct value, a wrong one and a missing field.
  • articles/PUBLISHING.md:
    • Step 1 now says review comes first and step 6 moves the article on to ready, after the author’s question.
    • Step 4 now mentions the published count and the repository’s own series lists, which this first publication showed were missing.
  • The metrics catch up: sandbox.yml had run after the merges of PRs #8 and #9 (11 s and 10 s, both identical), and neither run had been counted.

Commands that mattered

From the Makefile, and the check behind the new badge:

make article-check   # both articles at status published
make site-check      # fails if published: "N/7" disagrees with the statuses

Decisions

  • No ADR. The publication follows ADR-0007 and ADR-0008. The published field implements a badge already agreed in CLAUDE.md §9.3.
  • fig-grid-paths.png stays out of piece 1. The author asked why it wasn’t used, heard the reason and chose to leave the article as it is. The chart stays in the repository.

Dead ends and surprises

  • The checklist had a confusing first step. “The front matter says status: review” read as if ready were wrong, and the author stopped to ask mid-publication. A checklist is tested only when someone follows it for real.
  • No recorded reason for leaving out a chart. Session 08 used three of the four charts and didn’t say why this one went. The answer to the author had to be reconstructed, and was given as such: it mostly repeats the heatmap, which also says when.
  • Nothing here can read the published pages. Medium’s Cloudflare answers 403 to scripts. So the URLs are recorded as the author sent them, and the links on Medium weren’t checked from here.
  • A heredoc inside a heredoc. A script for site-check ended early, because both used the same delimiter. Nothing was written, and the edit went through the editor instead.

Who did what

  • The author: merged PR #10, published both articles, chose to leave the chart out, and sent the URLs.
  • Claude: created the tag, checked the links, added the badges and the published check, updated the repository to match Medium, and closed the session.

Next

  1. Example 2, the model-based reflex agent: internal state, and a run where that state survives stopping and restarting the cloud sandbox.
  2. Draft piece 2. The series goes out one piece a week (ADR-0007).
  3. The author: check that the two series links work on Medium, since nothing here can read those pages.