You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
docs content: links to objectstack-ai/spec — a second stale repository name — survive in two published blog posts and in docs.site.json #12366
The repository has been renamed at least twice, and links to both former names survive in the tree. All three names resolve to the same repository:
gh api repos/objectstack-ai/framework --jq .id → 1136691870
gh api repos/objectstack-ai/spec --jq .id → 1136691870
gh api repos/objectstack-ai/objectstack --jq .id → 1136691870
framework is being fixed at its single producer in #12354 / PR #12364. spec is a separate, unrelated set of spellings, and it is the one this card is about.
Where objectstack-ai/spec still appears
Renders to visitors today:
file
hits
content/blog/protocol-first-development.mdx
2
content/blog/metadata-driven-architecture.mdx
1
Inert config that spells a name nothing else spells — content/docs.site.json:
line 40 — themeConfig.socials[].url
line 63 — page.repoBaseUrl
Both have zero runtime consumers today: the only references to repoBaseUrl anywhere in the repo are gate scripts naming the file as a path-hint fixture. That is precisely why it is worth a card rather than a shrug — inert config carrying a wrong value is what someone wires up later, having reasonably assumed it was maintained.
Also present, and deliberately out of this card's scope unless the maintainer wants them in:CONTRIBUTING.md (3), docs/notes/airtable-dashboard-analysis.mdx, and the link-reference block at the bottom of the root CHANGELOG.md.
Prior art — read this before acting
docs/audits/2026-06-handwritten-docs-accuracy-followups.md:40 already flagged these links and declined to rewrite them. Whoever takes this card should read that entry and either follow its reasoning or say why it no longer applies — ⛔ do not re-litigate it from scratch, and do not silently overturn a recorded decision.
Why it matters
Every one of these costs a 301 today; none is broken. The reason to fix them is that a rename redirect is a signal that decays — GitHub keeps them only while the old name stays unclaimed — and because docs.site.json is a config file whose stale value is invisible until it is used.
Suggested scope
The two blog posts and docs.site.json. CONTRIBUTING.md, docs/notes/** and CHANGELOG.md are a maintainer call: CHANGELOG.md in particular is a historical record, and rewriting URLs inside past entries is not obviously right.
Source
Surfaced by the dev on #12354 while sweeping apps/docs/ for repository-name literals; the sweep was correctly held to apps/docs/ by its dispatch, and these hits are outside it. Every claim above was independently re-measured before filing.
Premise re-verified by symptom on origin/main before dispatch — 5 sites across 3 files, positive control: 68 files under content/ use the correct objectstack-ai/objectstack spelling, so the reads are instrument-true.
⭐ One site is worse than the card describes, and it changes the card's priority in kind rather than degree. The card frames this as stale prose in published text. docs.site.json:63 is repoBaseUrl — a functional configuration value, not prose. If it feeds the docs site's "edit this page" / "view source" links, then every such link on the whole site points at the wrong repository, which is a live user-facing defect rather than a cosmetic one. :40 is likely a nav/footer link with the same property.
Zone 2 for the dev, to falsify rather than inherit: I have not established what consumes repoBaseUrl. Read it out of the site code before deciding how to describe the fix — and if it turns out to be inert, say so with the evidence; that is a useful negative, not a failed dispatch.
⚠️ ⛔ Do not probe whether objectstack-ai/spec still exists by fetching it. This seat's GitHub scope is objectstack + objectui, and a 404 from outside that scope would be indistinguishable from a permissions refusal — it would be a reading your instrument cannot make. Decide the correct target from in-repo evidence (what the other 68 files spell, what the package metadata says).
Serial-queue check: no open PR claims either blog post or content/docs.site.json. Distinct from #12469 (api/client-sdk.mdx), #12451 (api/environment-routing.mdx) and #12433 (deployment/cli.mdx), all live in this lane right now.
Activity
yinlianghui commentedon Aug 26, 2026
Claim:
domain:devx @ objectstackexecution seat (#6023), sessionsession_01UjM2ia8Av1v5NqfqQEQmC6, round R11. Branchclaude/issue-12366-stale-spec-repo-name.Premise re-verified by symptom on
origin/mainbefore dispatch — 5 sites across 3 files, positive control: 68 files undercontent/use the correctobjectstack-ai/objectstackspelling, so the reads are instrument-true.⭐ One site is worse than the card describes, and it changes the card's priority in kind rather than degree. The card frames this as stale prose in published text.
docs.site.json:63isrepoBaseUrl— a functional configuration value, not prose. If it feeds the docs site's "edit this page" / "view source" links, then every such link on the whole site points at the wrong repository, which is a live user-facing defect rather than a cosmetic one.:40is likely a nav/footer link with the same property.Zone 2 for the dev, to falsify rather than inherit: I have not established what consumes
repoBaseUrl. Read it out of the site code before deciding how to describe the fix — and if it turns out to be inert, say so with the evidence; that is a useful negative, not a failed dispatch.objectstack-ai/specstill exists by fetching it. This seat's GitHub scope isobjectstack+objectui, and a 404 from outside that scope would be indistinguishable from a permissions refusal — it would be a reading your instrument cannot make. Decide the correct target from in-repo evidence (what the other 68 files spell, what the package metadata says).Serial-queue check: no open PR claims either blog post or
content/docs.site.json. Distinct from #12469 (api/client-sdk.mdx), #12451 (api/environment-routing.mdx) and #12433 (deployment/cli.mdx), all live in this lane right now.Generated by Claude Code