Another month, another pile of merged PRs to show for it. July’s headline stat is one I didn’t see coming: I gave 175 pull request reviews. Mind you, most of those reviews were of PRs written by robots, about changes suggested by robots. I’m not sure if I’m the manager or the middle manager at this point, but the org chart is definitely weird.
By the Numbers
| Metric | Count |
|---|---|
| Pull requests opened | 77 |
| … merged | 76 |
| … still open | 1 |
| … closed without merging | 0 |
| Issues opened (per my GitHub profile) | 9 |
| Commits (per my GitHub profile) | 113 |
| PR reviews given (per my GitHub profile) | 175 |
| Repos touched | 22 |
| Releases published | 7 |
| Bot-filed data updates | 0 |
Every PR this month had a human-shaped title — the daily data-bot took
July off, so that 77 is all me, or at least all me-supervised. A 76-of-77
merge rate with the one holdout being a PR to somebody else’s repo says
something about how agreeable my own repos are. The busiest repos were
this website (13), data-curated (11 — mostly The Tower game data
updates, because yes, I track a mobile game’s numbers like it’s my job),
and my-user-manuals (10).
Elsewhere
One PR went outside the family this month:
awesome-digitalocean #20,
adding fini-infra and dnscontrol to the configuration-management section
of the community’s awesome list and pruning a stale link while I was in
there. It’s still sitting open in their review queue, patient as a
barnacle. Upstreaming is a virtue; upstreaming slowly is apparently also
the community standard.
Releases Shipped
gh-observer v3.4, v3.5, and v3.6
Three gh-observer releases in ten days (v3.4 on July 4, v3.5 on July 9, v3.6 on July 13). The interesting one is v3.6: historical job-duration averages are now weighted toward recent runs, so the tool tells you what “normal” looks like this week, not what it looked like in a repo’s misspent youth. v3.4 and v3.5 were the dependency-and-housekeeping kind.
The rest of the fleet
template-repo v1.7 and v1.8 kept the shared justfile machinery sharp — including a nice one from v1.7: installing markdownlint on macOS without needing npm, which is the kind of quality-of-life fix nobody writes songs about but everybody appreciates. gh-amp v0.3 and fini-infra v4.3 — the latter adding zizmor for workflow security scanning and a new VPC — round out the month.
Notable Pull Requests
www-chicks-net #347: lighthouse to monitor web page quality
639 additions and 533 deletions — the biggest and most-discussed PR of the month with twelve comments of review cycles — wiring Lighthouse into CI to keep an eye on page quality automatically. A follow-up got the lighthouse runs going in parallel because waiting for sequential audits is for people with more patience than me. The site now grades itself in public, which is exactly the kind of accountability a blog needs and a blogger dreads.
my-user-manuals #47: meta.toml for the manuals
236 additions migrating the remaining user-manuals directories to
structured meta.toml front matter, complete with a CI check that
verifies the toml is correct. The manuals repo has quietly become one of
my favorite personal projects — documentation for hardware I own, from
band saws to 1990s modems, now with metadata strict enough to pass a
linter. Somewhere a librarian is smiling and doesn’t know why.
gh-amp #23: required-check status alongside mergeable
271 additions teaching gh-amp to show required-check status right next to mergeability, so you don’t have to guess which missing check is holding up the merge party. Small feature, outsized relief for anyone who has ever stared at a blocked PR wondering which check is the culprit.
Honorable mention to data-curated #390, “analyze dissonant runs” — 165 lines of analysis code written for a mobile game, which is the kind of thing that happens when a data engineer gets bored. No regrets.
Embedded Metadata
These are times where it seems good to be explicit about our choices about and usage of AI (Artificial Intelligence). Plus any links to cross-posts I do.
- This post was drafted by GLM-5.3 in OpenCode using the
gh-recapskill; the numbers came straight from GitHub’s API via theghCLI on 2026-10-30. The review-cycle counts come from the actual PR comment threads, robots and all. - The PR carrying this series was code-reviewed in CI by Claude Code — robots reviewing robots, which felt thematically correct.
- The draft was reviewed and edited by me before publishing. The opinions, jokes, and any errors are mine.