Skip to content

notes: the sweeps ran — all four gone, and what the live run settled - #377

Merged
JulienMartel merged 1 commit into
mainfrom
worktree-peppy-meandering-dragonfly-sweep-ran
Aug 15, 2026
Merged

notes: the sweeps ran — all four gone, and what the live run settled#377
JulienMartel merged 1 commit into
mainfrom
worktree-peppy-meandering-dragonfly-sweep-ran

Conversation

@JulienMartel

Copy link
Copy Markdown
Contributor

What

A notes-only follow-up to #375. The handoff and §5.3 both said "until both have run, two of the four are still public" — both have run, so that line is now false in the plan of record.

Why

Both preview-sweep copies merged and ran 2026-08-15 11:16 UTC, dry-run first then for real. All four leaked Workers deleted. Re-enumerating the account leaves nebelhaus, hausfold and hausfold-pr-49, and the two hostnames that were serving /init.sh?ref=<sha> now 404 — so the ?ref= hole is gone from the code's last hiding place, not just from the zone.

Three things the live run settled that construction alone could not, and they're why this is worth a commit rather than a mental note:

  • The keep arm is real, not just anchored. hausfold.co's run had an open PR's preview (hausfold-pr-49) in front of it and left it alone, in both passes. That's the safety property that most needed demonstrating before anyone trusts this on a cron.
  • The CI token can list. The header called Workers Scripts:Read "inference until the first run". It isn't any more, in either repo.
  • preview-web.yml's fast path is fine when it fires. ci, web: enumerate the account, reap leaked previews, and fix three 301s onto a 404 #375's own preview was already gone before the sweep looked — backstop, not replacement, and the account listing is what shows the difference.

Verify

workshop      DELETED nebelhaus-pr-321 / nebelhaus-pr-341
hausfold.co   DELETED hausfold-pr-16 / hausfold-pr-22, keep hausfold-pr-49 (PR #49 is open)

Then, against the account and the live URLs:

# the two that served the hole
curl -s -o /dev/null -w '%{http_code}\n' \
  "https://nebelhaus-pr-341.<sub>.workers.dev/init.sh?ref=<a resolvable sha>"   # 404

# untouched
curl -sI https://hausfold.co/                      # 200
curl -sI https://nebelhaus.com/guides/pounce       # 301 → /docs/haus/rooms/launcher/

Watch out

🤖 Generated with Claude Code

https://claude.ai/code/session_01PkTBtapKcZTCeTqfhRZEJ7

Both preview-sweep copies merged and ran 2026-08-15 11:16 UTC, dry-run first
then for real. All four leaked Workers deleted; re-enumerating the account
leaves nebelhaus, hausfold and one open PR's preview, and the two hostnames
that were serving /init.sh?ref=<sha> now 404 — checked against the account and
the URLs, not read off the job log.

Three things the live run settled that construction alone could not. The keep
arm is real: hausfold.co's run had hausfold-pr-49 (an open PR) in front of it
and left it alone in both passes, which is the safety property that most needed
demonstrating before anyone trusts this on a cron. The CI token can list — the
header called Workers Scripts:Read "inference until the first run", and it
isn't any more. And preview-web.yml's fast path is fine when it fires: #375's
own preview was already gone before the sweep looked, which is the difference
between a backstop and a replacement.

Production was untouched: hausfold.co 200, nebelhaus.com/guides/pounce still
301s onto the launcher room.

Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PkTBtapKcZTCeTqfhRZEJ7
@JulienMartel
JulienMartel merged commit 25a3040 into main Aug 15, 2026
4 checks passed
@JulienMartel
JulienMartel deleted the worktree-peppy-meandering-dragonfly-sweep-ran branch August 15, 2026 11:23
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant