chore: retire GitHub references across docs, links, and workflows #26
No reviewers
Labels
No labels
dependencies
major
security-pin
No milestone
No project
No assignees
1 participant
Notifications
Due date
No due date set.
Dependencies
No dependencies set
Reference
trails-cool/trails!26
Loading…
Reference in a new issue
No description provided.
Delete branch "chore/retire-github-references"
Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?
GitHub was retired on 2026-07-27, but 21 files still pointed at it. Two were broken in production.
The 404s
Deleting
mainon GitHub broke two user-facing paths:github.com/trails-cool/trails/blob/main/SECURITY.mdraw.githubusercontent.com/…/main/infrastructure/docker-compose.ymlSo someone following our own security-contact path hit a dead page, and anyone following the install instructions failed at step one.
Both now point at
git.ullrich.is. Because these are public-facing, I verified them anonymously with no token — existence isn't the requirement, reachability without credentials is:Also repointed: journal footer + planner landing "source" links (these resolved to the
movedstub), SECURITY.md scope, the BRouter host runbook's clone step, anddocs/architecture.md, which still called CI/CD "GitHub Actions" and named the GitHub org as the repo's home.CLAUDE.md documented a workflow that cannot work
The Git Workflow section instructed
gh pr list,gh pr create,gh pr merge --merge --auto, and a GitHub merge queue.ghdoes not talk to Forgejo — every command in that section failed. Replaced with Forgejo API equivalents, and stated plainly that there is no merge queue (#25 tracks the evaluation), including the consequence that matters: PRs are tested against their base at push time, not merge time, which is the real risk when landing stacked work quickly.Two dead conditions, handled differently on purpose
ci.ymlgated Gitleaks ongithub.actor != 'dependabot[bot]'. That existed because Dependabot PRs ran without secret access, soGITLEAKS_LICENSEwas empty and the scan failed. Dependabot cannot exist on Forgejo and Renovate runs with full secret access — so the gate is removed, not ported. A dependency bump is exactly when a secret can slip into a lockfile; that scan should run.cd-staging.ymlexcludeddependabot/github_actions/head refs from preview deploys. That intent still applies to Renovate, so it is ported torenovate/rather than deleted.Deliberately kept
.github/copilot-instructions.md— I flagged this as vestigial earlier and was wrong. It is a 3-line pointer toCLAUDE.md; Copilot reads it in-editor independent of GitHub hosting, and its content is accurate. Deleting it would break something still working.openspec/changes/archive/**anddocs/conversations/**— historical record. Rewriting them would falsify what happened.uses: https://github.com/trails-cool/gitleaks-action@<sha>— a live dependency on a GitHub-hosted fork under our own org. Verified both the repo and the pinned SHA still resolve (200). Worth knowing that "GitHub is retired" is not quite complete while CI resolves an action from there; mirroring it to Forgejo is a real follow-up, not a search-and-replace.Verification
pnpm typecheck14/14 andpnpm lint14/14 pass.pnpm testneeded two runs to be trustworthy: the first reported 5 packages failing, the re-run passed 12/12 clean with no code change between them, and@trails-cool/fitpassed in isolation — contention from running the three suites back-to-back locally, not this diff. Also confirmed no test or e2e spec asserts any of the changed hrefs, so nothing was silently coupled to the old URLs.Memories were updated in the same sweep (outside the repo), most importantly one that told future sessions "GitHub kept as mirror — do NOT remove anything from GitHub" — now struck through and superseded.