HTML Recovery Tool: Restore lecture sites from GitHub Release assets

Producer live: lecture-python.myst, lecture-python-intro, lecture-python-advanced.myst, lecture-python-programming, lecture-python-programming.fr, lecture-jax and lecture-dp pass create-release-assets: 'true' to publish-gh-pages, and each latest release carries the triple — tarball, SHA256 checksum, JSON manifest (name, tag, commit, timestamp, size_mb, file_count, repository). The only publish-gh-pages/action.yml change from v0.8.0 to HEAD is an action-gh-release pin bump, so asset shape is identical across the pins those seven use (v0, v0.8.0, v0.9.0). Missing is the consumer; built HTML is costly to reproduce, so a downed site recovers fastest from the archive.

Scope

• Download latest release asset (HTML tar.gz) from any lecture repo
• Verify integrity via the SHA256 checksum asset
• Extract and optionally serve locally for verification
• Deploy to GitHub Pages or alternative hosting
• Batch recovery across multiple repos
• List available recovery points — this is just gh release list

Gaps

Two asset formats. lecture-julia.myst, lecture-python.zh-cn, lecture-intro.zh-cn, lecture-python-programming.fa hand-roll the triple in their own publish.yml (tar + sha256sum + heredoc manifest, uploaded via softprops/action-gh-release directly, @v2 in the first, @v3 in the rest): same <name>-html-<tag>.tar.gz, but checksum and manifest are the bare html-checksum.txt / html-manifest.json with no asset-name prefix, and the manifest omits name and repository. Handle both, or migrate those four onto publish-gh-pages first.

Tag-triggered publishes only. publish-gh-pages/action.yml:90-94 skips off-tag with a warning, and lecture-dp also accepts workflow_dispatch (publish.yml:6), so a manual publish there leaves no recovery point. Nine active repos attach nothing — lecture-datascience.myst, lecture-dle, lecture-dps, lecture-dynamics, lecture-eqm, lecture-stats, lecture-tools-techniques, lecture-python-programming.ml, lecture-python-programming.zh-cn deploy via peaceiris/actions-gh-pages or actions/deploy-pages with no archive step, outside recovery until they adopt create-release-assets.

Checksum/manifest names carry no tag. publish-gh-pages/action.yml:116-117 builds both from ASSET_NAME alone while the archive at :108 includes it, so one repo's recovery points collide in a directory — namespace by repo and tag.

Where it lives

PLAN.md:116: "Decide home, then build". workflow-backups already has src/backup/repo_matcher.py for --org/--pattern selection, plus S3 and tests; if it goes there, this repo only keeps the producer contract stable.

• Decide the home — transfer to workflow-backups (worth timing against QuantEcon/workflow-backups#14, which proposes renaming that repo) or build here
• Either way, state in publish-gh-pages/README.md (asset list at :89-92) that the triple and its naming are a compatibility contract for recovery — can ride #109

Related

• Producer publish-gh-pages/action.yml:80-154 (create-release-assets): guards :90-98, naming :100-108, checksum :123, manifest :126-136, upload via softprops/action-gh-release SHA-pinned to v3.0.2 at :147 (#101)
• publish-gh-pages/README.md:76-92 describes the triple; workflow-backups (S3 git mirror backups) is the likely destination

Start with PLAN.md:116 and compare building here with transferring the work to workflow-backups, especially src/backup/repo_matcher.py, its S3 integration, and tests. Read publish-gh-pages/action.yml:80-154 and README.md:76-92 to confirm both release-asset formats and their naming; done means the home is decided, the producer contract is documented, and recovery can verify, extract, and deploy supported archives.