Add on-demand GitHub Pages previews of decks changed in pull requests

[CodyCBakerPhD]

Manually triggered previews of only the decks a PR adds/changes (e.g. 2026-dhmc-ieeg.html in #26), published to gh-pages like dandi/usage-page does.

How it works

.github/workflows/preview.yml runs only on workflow_dispatch (Actions → Preview → Run workflow) with two inputs: a PR number, and deploy or remove.

• build-preview job (read-only token)
  • checks the PR is open and checks out its merge ref refs/pull/<N>/merge;
  • finds top-level *.html files added/modified/renamed by the PR (git diff HEAD^1 HEAD), and fails clearly if there are none;
  • inits only the reveal.js + reveal.js-mermaid-plugin submodules (not 2022-nih-compcore, whose relative URL only resolves on datasets.datalad.org);
  • installs git-annex, fetches origin/git-annex, adds https://datasets.datalad.org/centerforopenneuroscience/talks/.git as a remote and git annex inits;
  • runs tools/build_preview.py, taken from the workflow's own commit (so the PR being previewed doesn't need to contain it), against the PR checkout. It walks the deck's local references (src, data-src, data-background*, data-markdown, stylesheet hrefs, markdown images, CSS url()), recursing through HTML/CSS/MD, runs git annex get on anything without content, and copies just those files (dereferenced) into the site;
  • content it could not get (e.g. not yet pushed to datasets.datalad.org) is listed in the job summary and as warnings; the build does not fail.
• deploy-preview job (write token; receives only the built files)
  • deploys with rossjrw/pr-preview-action to pr-preview/pr-<N>/ on gh-pages, or removes that directory;
  • posts or updates a single comment on the PR linking each changed deck directly.

Setup (done)

• Pages is served from gh-pages / root; that branch has a placeholder index.html and .nojekyll.
• Actions workflow permissions: Read and write.
• master already has a copy of the workflow's opening section, so "Run workflow" is available. Until this is merged, select this branch under "Use workflow from".

Testing

• actionlint clean.
• Run 37684981023, dispatched from this branch for 2026-dhmc-ieeg: NWB/NeuroConv + Neurosift event-related slides #27: deck detection found 2026-dhmc-ieeg.html, git annex get fetched 8 files from datasets.datalad.org, 43 files were deployed, and the bot posted the preview link on 2026-dhmc-ieeg: NWB/NeuroConv + Neurosift event-related slides #27 → https://con.github.io/talks/pr-preview/pr-27/2026-dhmc-ieeg.html. Four images from Slide for the 2026 dhmc ieeg group meetup this Fri #26 whose content isn't on datasets.datalad.org yet were listed as missing, as intended.

Known trade-off

Each deploy commits the preview's files to gh-pages, so its history keeps growing (about 13 MB for that run). See the discussion below.

🤖 Generated with Claude Code

https://claude.ai/code/session_01WaJ8d7w97BG5hbJVRr9B3M

[CodyCBakerPhD]

I get a bunch of interoperability issues when trying to build some of these slide decks through npm (even when build is successful) - would be nice to have centralized, stable builder to serve and be convenience to preview the deployment prior to moving to its final destination

@yarikoptic Can you go to settings -> pages -> enable pages from branch (gh-pages)?

[CodyCBakerPhD]

(or give me admin for the day so I can fanangle with it)

[yarikoptic-gitmate]

@yarikoptic here: made you admin here BUT note that we crossed in the internet -- I was just moving this repo over to https://hub.centerforopenneuroscience.org/con/talks . Where eventually we would deploy https://codeberg.org/git-pages/git-pages to fascilitate gh-pages like behaviors ... "migration" even moved issues and PRs ... we could for now continue here, and potentially re-migrate later etc if so desired I guess.

[yarikoptic-gitmate]

note that it might become 'prohibitive' after committed directly to git images exceed ultimate size allowed etc. I think I ran into this with annextube demos etc... overall -- I feel that it might be a bit of a wasted effort , but you could be proven me wrong

[CodyCBakerPhD]

note that it might become 'prohibitive' after committed directly to git images exceed ultimate size allowed etc. I think I ran into this with annextube demos etc... overall -- I feel that it might be a bit of a wasted effort , but you could be proven me wrong

@yarikoptic-gitmate I am a tad confused since that sounds like the real @yarikoptic but I was told that @yarikoptic-gitmate was only for "AI-produced" output (hence the entirely separate account?)

Anway it's working great: https://con.github.io/talks/pr-preview/pr-27/2026-dhmc-ieeg.html#/4/6

If no qualms I will merge this later

[yarikoptic]

yeah, it was me just in a browser where I had the mate's session, so I started the post with "@yarikoptic here:" to disambiguate.

Greate that it works, I glanced over and looks minimalistic enough which is pretty cool -- let's then get it for now, thanks!