[d-morrison]

Closes #30
Closes #35

Course sites no longer include mds as a git submodule: mlds has no .gitmodules and links to these pages by URL. The rule against adding Quarto extensions assumed that every host site would have to install them, so it no longer applies.

• Removes the extension rule from CLAUDE.md and from .github/instructions/r-and-config.instructions.md.
• Rewrites the project-context sentence: page paths and #id anchors are still an interface, because other sites link to them by URL.

Not in this PR: the index.qmd "Using these notes in another site" section and the self-symlink notes still describe the submodule setup. They are tracked in #31.

Also removes the same rule and framing from the review bot's prompt in .github/workflows/claude-code-review.yml and from a comment in copilot-setup-steps.yml (#35).

🤖 Generated with Claude Code

[d-morrison]

Deferred: the stale rule in .github/workflows/claude-code-review.yml's prompt addendum (and the "downstream books" comment in copilot-setup-steps.yml) is tracked in #35.

It's deferred rather than fixed here because editing the review workflow in this PR can make the review bot skip reviewing it, which would leave this PR without a verdict. That change goes in its own PR. Every copy of the rule outside .github/workflows/ is removed in this PR.

No code changed this round.

@claude review

Posted by Claude Code (AI agent) --- not written by a human.

[github-actions]

👀 Picked up by workflow run #36499118451. Setup runs first; Claude itself responds after that.

[github-actions]

Claude finished review — View run

---

YAML is valid and the block renders cleanly (item 2 now reads as a single bullet, no dangling numbering or indentation issues). Everything checks out — all three previously-flagged findings are resolved and no new ones were introduced by this round's changes.

Review: PR #32 — Drop the no-new-extensions rule and the stale submodule premise

What changed since the last round: The prior round reviewed b5baa01376c1bec28c57f1f08088f3a5954da0c5; the computed range to the current head af4b8865658cd149350581b9cc09a1ebd2747e8c is:

af4b886 Merge 075277ad51c50fc9fcee17e494735e774d722c78 into 9b639b36659ce6f853bfb5e06b7711fb55580697
075277a Merge branch 'main' into claude/claude-md-extensions
9b639b3 Style remarks like definitions: minimal callout, not foldable (#36)
72e07d8 Define strict subset, superset, and strict superset (#29)
d41beeb Add a Measures page, moved from the probability notes (#34)

All five commits are a merge of already-merged, unrelated main-branch PRs (#29, #34, #36 — new content pages on measures and set-relation definitions). I confirmed with git diff b5baa013 af4b8865 -- CLAUDE.md .github/instructions/r-and-config.instructions.md .github/prompts/quarto-review.prompt.md that this range is empty for this PR's own files — none of the merged content touches anything in this PR's actual scope. gh pr view 32 confirms changedFiles: 5, +9/-17, matching .claude-review-pr.diff exactly.

Every finding raised in the three prior rounds is now resolved:

1. .github/pkg.lock (flagged round 1) — removed by 75211e6, confirmed absent from the current diff and from HEAD.
2. .github/workflows/claude-code-review.yml:75-76 downstream-book dependency rule (flagged rounds 1–3) — removed by b5baa01. Verified the merged file (git show HEAD:...) parses as valid YAML, the "2. R code" section now reads as a single clean bullet with correct indentation, and the "4. CI / template impact" heading was correctly retitled to "4. CI impact" to match the retired template framing.
3. .github/instructions/r-and-config.instructions.md and .github/prompts/quarto-review.prompt.md — dependency rule removed, quarto-review.prompt.md's checklist correctly renumbered 1–7 (was 1–8) with no gaps or duplicate numbers.
4. CLAUDE.md — "Project context" now bases the path/#id-anchor interface on mlds linking by URL rather than the retired submodule premise, and "Things to avoid" no longer lists the dependency rule.
5. .github/workflows/copilot-setup-steps.yml — the "downstream books copy" comment fragment was dropped, consistent with the rest of the cleanup.

I also did a repo-wide git grep at HEAD for "downstream book", "every dependency", "new top-level dependencies", and "new R package or Quarto extension" (including .github/copilot-instructions.md) and found zero remaining hits — the rule and its submodule/template rationale are fully retired everywhere, not just in the files originally diffed.

Note: this checkout's on-disk working tree shows CLAUDE.md and the two workflow files as locally modified/reverted (per git status, and a .github/workflows/.restored-from-default-branch marker) — that's the review harness's default-branch-fallback and stale-worktree artifacts, not part of the PR. git show HEAD:<path> confirms the actual committed content matches the diff above in all cases, so this doesn't affect the assessment.

No new issues found. This confirms and closes out the "Needs more work" verdict from the prior round (2026-09-28T23:49:26Z summary) — that round's sole outstanding item (claude-code-review.yml:74-76) is now fixed, and no regressions were introduced by the intervening merge of unrelated main-branch PRs.

Verdict

Ready for merge

Structured Review Data (JSON)

{
  "schema_version": "1.1",
  "reviewer": "claude",
  "commit_sha": "af4b8865658cd149350581b9cc09a1ebd2747e8c",
  "verdict": "CLEAN",
  "findings": [],
  "detailed_assessment": "No detailed findings: git diff b5baa013..af4b8865 restricted to the PR's own files (CLAUDE.md, r-and-config.instructions.md, quarto-review.prompt.md) is empty, and the intervening merge only pulled in unrelated already-merged main-branch content pages.",
  "holistic_assessment": "All three prior rounds' findings (stray pkg.lock, stale downstream-book dependency rule in claude-code-review.yml) are now fixed and a repo-wide grep confirms zero remaining references, satisfying the PR's full stated scope."
}

Reviewed commit: 075277a

[github-actions]

💰 Cost: $0.3543 (review) — run

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-09-28 18:00 PDT