[cscheid]

Fixes bd-cvk7jb11, which turned up during bd-73lilznv (#929).

Commands that wrote to the main checkout from a worktree

create_worktree::repo_root() returns the main checkout, the parent of --git-common-dir. That is the right root for placing new worktrees, but four commands also used it to find the files they work on. Run from .worktrees/<x>/, each of them read or wrote the main checkout instead:

• stage-doc-examples rendered the main checkout's examples/ and staged into its docs/examples/. The worktree's own q2 render docs then failed with "Declared resource …/docs/examples does not exist on disk".
• check-noop-rebuild built the main checkout's q2.
• gen-math-spec read and rewrote the main checkout's commands.json.
• braid-snapshot wrote the main checkout's .braid/snapshot.jsonl.

They now use switch_task::current_worktree_root() (--show-toplevel). The capture-*, pandoc-check, render-corpus-diff and pull-extension-subtree commands already used it. repo_root() is renamed main_checkout_root() so the next caller can see what it returns. Its two remaining callers, create-worktree and render-corpus-diff's reference copy, want the main checkout on purpose.

switch-task --from <branch> branched off a stale tip

It fetched origin/<branch> but then branched off the local <branch>, while printing "after pull". The local branch is usually checked out, and stale, in another worktree; main in the main checkout is the common case. It now chooses the start point like this:

• It uses whichever of the local branch and origin/<branch> is newer.
• It stops if the two have diverged.
• It passes --no-track, so the topic branch never tracks the integration branch.

.claude/rules/worktrees.md is updated to match.

Excluded crates walked up into the main checkout's workspace

cargo finds a crate's workspace by walking up parent directories. It skips, rather than stops at, a workspace that excludes the crate. In a worktree under .worktrees/, the walk therefore continues past the worktree's root Cargo.toml into the main checkout's, which doesn't list the crate. The three excluded crates without a [workspace] of their own were affected:

• crates/wasm-qmd-parser could not be built from a worktree at all.
• crates/pampa/fuzz and crates/tree-sitter-language-wasm-shim failed cargo metadata. This broke check_feature_sets::discovers_the_wasm_consumers_in_this_repo in every worktree.

Each now declares an empty [workspace]. That is what cargo's error message suggests and what wasm-quarto-hub-client already does for the same reason. A note at the root exclude list states the rule. Nothing changes in the main checkout or CI.

Verification

• Run from a worktree, cargo xtask stage-doc-examples now stages into that worktree, and the main checkout stays untouched and clean.
• 6 new tests cover the start-point choice (behind, equal, ahead, diverged, local-only and remote-only) and check that the two root helpers differ in a linked worktree.
• cargo nextest run -p xtask passes 250/250 from a worktree; it was 249/250 before the [workspace] change.
• From the worktree, cargo metadata succeeds for all three excluded crates and npm run build:wasm passes; hub-client builds through the shim. wasm-qmd-parser now gets past workspace lookup. Its full wasm-pack build was not run locally; CI's WASM jobs cover it.
• cargo clippy -p xtask --all-targets -- -D warnings passes on the host and with --target x86_64-pc-windows-msvc.
• q2 render --strict claude-notes is clean.

🤖 Generated with Claude Code

[posit-snyk-bot]

✅ Snyk checks have passed. No issues have been found so far.

Status	Scan Engine	Critical	High	Medium	Low	Total (0)
✅	Open Source Security	0	0	0	0	0 issues
✅	Licenses	0	0	0	0	0 issues

💻 Catch issues earlier using the plugins for VS Code, JetBrains IDEs, Visual Studio, and Eclipse.