[devin-ai-integration]

What and why

Follow-up to #1833. The server offered Run/Evaluate code lenses on documents the VS Code runner cannot read: run() in editors/vscode/src/run.ts refuses any lens whose uri is not a file: URI ("… is not declared in a file on disk, so sysml cannot read it."), so on those documents every Run/Evaluate lens was a button that could only fail.

• internal/frontend/lsp/codelens.go CodeLens: the runCommand lens is emitted only when params.TextDocument.URI has the file scheme (onDisk). Reference-count lenses are unchanged.
• internal/frontend/lsp/notebook_requests.go cellRequest (textDocument/codeLens): drops lenses whose Command.Command == runElementCommand. The kernel runs a cell, so a cell gets only its reference-count lenses, which still resolve through codeLens/resolve in the cell's terms.
• run.ts keeps its client-side check unchanged, as a defensive fallback.

Which guard removes the cell lenses. Guard 1 alone does not cover cells. cellRequest calls CodeLens with nameToURI(nb.name), which is uri.File(...) of the notebook's path, so it is always a file: URI. With only guard 1 applied, the new notebook test failed: the cell's answer still had Run Go / Evaluate Area lenses with "uri":"file:///…/model.ipynb". The runElementCommand filter in cellRequest is what removes them, so it is not redundant.

Which URIs reach CodeLens from VS Code. editors/vscode/src/selector.ts documentSelector registers {scheme: "file"}, {scheme: stdlibScheme} (library documents, which CodeLens already answers with nothing) and {notebook: "*"} cells. It does not include untitled, so from VS Code an untitled editor never reaches the server, and a cell's request reaches CodeLens through cellRequest with the notebook's file: URI. The scheme guard is still right for other LSP clients that send non-file URIs (untitled:, virtual file systems), and the new test locks it in with untitled:Untitled-1.

Docs: docs/reference/lsp.md (the codeLens-on-a-cell sentence), docs/guide/12-jupyter.md (what a cell has), docs/guide/08-editors.md (the running-an-element paragraph and the code-lens feature row) and editors/vscode/README.md (notebook cells, "Running an element from a code lens") now say that Run/Evaluate lenses are offered only for model files on disk, that the kernel runs a cell, and that reference-count lenses work in cells. Changelog: changes/unreleased/notebook-run-lenses.fixed.md.

How it was verified

New tests in internal/frontend/lsp:

• TestNotebookCellLensesCountReferencesAndRunNothing (notebook_test.go): a cell with part def A, action def Go and calc def Area, plus a second cell using them. The cell's textDocument/codeLens answer has reference lenses on P::A, P::Go and P::Area and no opensysml.runElement lens, and codeLens/resolve on P::A gives 1 reference located in the second cell. It fails with guard 1 alone (see above).
• TestCodeLensRunsOnlyFilesOnDisk (editor_facts_test.go): editorFactsModel opened as untitled:Untitled-1 gets reference lenses on Demo::Fall, Demo::Rover, Demo::Charge, Demo::Mission and Demo::Positive, and no command lens. With the guard forced off it fails (lens "Evaluate Fall" (opensysml.runElement) on a document not on disk, …).
• The existing file-document tests (TestCodeLensOffersReferenceCountsAndRunCommands, TestCodeLensResolveCountsReferences, TestCodeLensTellsOverloadsApart, TestRunServesSignatureHelpCodeLensAndInlayHints) are unchanged and pass, so a file: document still gets Run/Evaluate.

$ go build ./...                      # ok
$ go vet ./...                        # ok
$ gofmt -l .                          # (empty)
$ go test ./internal/frontend/...     # all ok (lsp 26.2s, repl 73.7s, jupyter, mcp/*, stdiorpc, …)
$ go test -count=1 ./tests/hygiene/...
ok  	github.com/Open-MBEE/OpenSysML/tests/hygiene	14.185s
$ go test -race -count=3 -run 'TestCodeLens|TestNotebook' ./internal/frontend/lsp
ok  	github.com/Open-MBEE/OpenSysML/internal/frontend/lsp	2.323s
$ make lint                           ✓ Lint passed (staticcheck + gosec)
$ go run -C tools ./cmd/doc-counts -check
doc-counts: already current
$ python3 scripts/check-doc-links.py
0 broken link(s)

No TypeScript was touched, so the extension's npm test was not run.

Checklist

• make lint passes locally; the touched packages' tests pass (go test ./internal/frontend/..., ./tests/hygiene/...); the full make test is left to CI
• Tests added or updated for the change
• Documentation extended where it already covers the surface (see CONTRIBUTING.md)
• Changelog entry added as changes/unreleased/<slug>.<section>.md, not as an edit to CHANGELOG.md
• baselines regenerated and make docs-counts run if a gate count moved (compliance rows need nothing: the census is counted at docs build). No gate count moved (doc-counts -check is current)
• No internal work-item labels (waves, slices, F4, K5) in the body, docs, or changelog

[devin-ai-integration]

I'll fix CI failures and address comments from users with write access. I'll skip comments containing "(aside)".

• Disable automatic comment, CI, and merge conflict monitoring

[devin-ai-integration]

✅ Devin Review: No Issues Found

Devin Review analyzed this PR and found no bugs or issues to report.