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
CLAUDE.md says this repo is private; it is public #86
CLAUDE.md stops asserting a repo visibility that is the opposite of the truth,
so nobody reasons from it the way I just did.
What happens if we never do it
The line reads as licence to put internal-only content in this repo. Nothing
would flag it, and by the time it mattered the content would already be public —
in the repo and on the pkgdown site, which serves CLAUDE.md and CLAUDE.html
verbatim and indexes them in search.json.
The stale claim
CLAUDE.md line 8:
- **Origin:** `NewGraphEnvironment/cd` (private)
gh repo view --json isPrivate returns false. The repo is public.
Found 2026-09-07 while checking .Rbuildignore during #84. I read the line,
treated it as fact, and started writing up a published-internal-content leak
before checking — which is the exact failure the line invites.
The second half, lower stakes
.Rbuildignore carries only LICENSE.md, _pkgdown.yml, docs, pkgdown, .github. So planning/ and CLAUDE.md ship in the built tarball, and pkgdown
publishes CLAUDE.md (7,723 lines of agent instructions) as a page plus a
verbatim .md copy plus full-text search entries:
Because the repo is public this discloses nothing new — it is untidy, not a leak,
and worth separating from the line above so the two are not conflated. The remedy
is the one in the pkgdown-publishing convention: remove CLAUDE.md in CI beforebuild_site() (deleting docs/CLAUDE.html afterwards leaves the .md
and the search index), and gate on a declared allowlist so a new root markdown
file cannot appear silently.
Suggested fix
Correct the visibility line — and check the rest of the header block against
reality while there.
Add ^planning$, ^CLAUDE\.md$, ^scripts$, ^data-raw$ to .Rbuildignore
as appropriate, then verify against the tarball rather than the config: R CMD build . >/dev/null && tar tzf cd_*.tar.gz | grep -c '^cd/planning/'
Add the pre-build removal step and allowlist gate to pkgdown.yaml, and test
the gate against both known answers before trusting it.
First half done on main: the visibility line now reads (public) and points at gh repo view --json isPrivate so the same staleness is cheap to re-check.
Leaving this open for the second half, which is a real change rather than a doc fix: planning/ and CLAUDE.md still ship in the tarball and onto the pkgdown site. Because the repo is public that discloses nothing, so it is untidy rather than urgent — but the .Rbuildignore additions want verifying against the built tarball, and the pkgdown allowlist gate wants testing against both known answers, neither of which belongs in a drive-by commit.
Activity
Correct the repo visibility claim in CLAUDE.md
NewGraphEnvironment commentedon Sep 7, 2026
First half done on main: the visibility line now reads
(public)and points atgh repo view --json isPrivateso the same staleness is cheap to re-check.Leaving this open for the second half, which is a real change rather than a doc fix:
planning/andCLAUDE.mdstill ship in the tarball and onto the pkgdown site. Because the repo is public that discloses nothing, so it is untidy rather than urgent — but the.Rbuildignoreadditions want verifying against the built tarball, and the pkgdown allowlist gate wants testing against both known answers, neither of which belongs in a drive-by commit.