The ask
I asked chatgpt to build a Featured Projects page for Round 1 of the harness bake-off, covering the American Film Market IT operation, my MSP team’s AI tooling and this AI-managed WordPress site. The entry had to use existing sources, core blocks and the active theme’s presets.
What changed
- Published Featured Projects, the chatgpt Round 1 entry, as a child of the round page, with the required title and slug.
- Gave each of the three projects a context, my role, tools and approach, and a concrete explanation of why it matters.
- Used matching project sections, descriptive source links and jump links. Core columns place role and tools side by side on desktop and stack them on phones.
- Kept one H1 from the page title and used H2 and H3 headings in order. All added color, spacing and font-size choices use theme presets.
How
I used chatgpt over Tailscale SSH, with HARNESS=chatgpt on every ops command. It pulled the repo and read all four required rule files before opening the run. Facts came from Home, About, Experience, AI & Automation, the GitHub tracking Lab Note, README and the shared guidelines. A separate snapshot preceded publication through ops/wp.sh, and the verified entry was committed and pushed through ops/sync.sh.
Verification included a WP-CLI readback of page 89, HTTP 200 for the entry and its three source links, and desktop and 390-pixel phone previews in the in-app browser. The rendered page has one H1, all jump targets exist, columns stack on the phone, and the phone preview has no horizontal overflow. The entry commit changed only its page exports plus the generated index, database export and action log.
What worked, what didn’t
- Worked: the source material supported all three stories without adding new numbers, clients or measured performance claims. Core blocks provided the desktop and phone layout without custom CSS.
- Connection limit: the initial SSH hostname lookup failed until network access was granted. The earlier interrupted start did not leave an open run; the resumed start opened this run normally.
- Preview limit: Edge was unavailable to the browser tool. The in-app browser successfully loaded the entry and supported desktop and phone checks.
- Deliberate omission: I skipped images because the available stock photography would not document the actual projects. I did not guess technology brands for the AFM build or claim time savings for the AI tooling.
- No rollback was needed. I did not read competing entries or their bake-off Lab Notes, and did not manually change other pages, menus, settings, templates, themes, plugins or media. The required Lab Note script handles its automatic scoreboard refresh.
Change record
| Harness | chatgpt |
| Date | 2026-10-09 17:28 |
| Latest snapshot | 20261009-172753 |
| Undo this session | ops/rollback.sh --git 6ce336b |
Commits