fix(repo): honor --skipGitInit in both create paths

Why

repo create --skipGitInit (-s) documents itself as "Skip local repository setup (skips both git init and cloning)." Two paths did not honour it, and they fail differently depending on whether a project name is passed. Both are described in #8533 (closed), with an implementation guide from @phikai that this follows.

Bare -s in a non-Git directory runs git remote add anyway, and exits 0.

$ mkdir /tmp/example && cd /tmp/example && glab repo create -s
✓ Created project on GitLab: ... - https://gitlab.com/phikai/example
fatal: not a git repository (or any of the parent directories): .git
Warning: Could not add remote: exit status 128
$ echo $?
0

With no positional argument isPath is true, and -s short-circuits the git-init check, so needsGitInit stays false. Control reaches the else that calls addRemote with no check that the directory is a repository at all. The failure is reported with LogErrorf and the function returns nil, so a script checking the exit status sees success.

-s with a project name still prompts for, and performs, local setup.

With len(args) == 1 the command leaves isPath false, so the isPath block is skipped and control reaches the local-setup branch — where skipGitInit is read but never consulted. On a TTY it prompts Create a local project directory for <namespace>/<project>? and, if accepted, drives repoInitializer/repoCloner. That is the cloning the flag says it skips.

What

• the remote is added only when isGitInitialized, which was already computed a few lines up
• the local-setup condition gains !skipGitInit

The first also covers the interactive case where the user declines git init: that path reaches the same else, and there is no repository for a remote to go into either way.

Verification

Both tests t.Chdir into an empty temp directory, so git rev-parse --git-dir fails and isGitInitialized is false — the reported setup rather than an approximation of it.

check	result
go test ./internal/commands/project/...	ok, all packages
go vet ./internal/commands/project/create/...	clean
make gen-docs then git status	clean, so check_docs_update has nothing to report

Mutation-tested, and the second one changed the test because of what the mutation showed:

mutation	result
revert the isGitInitialized guard	FAIL — does not touch the remote, in 3s
revert !skipGitInit	FAIL — no prompt and no local setup, in 10s

The second mutation originally hung rather than failing: the reintroduced bug prompts, and the confirm blocked forever on a TTY nobody is typing into. A test whose failure mode is a timed-out CI job is not much of a regression test, so it now runs the command under a context deadline — with the flag honoured it finishes in milliseconds, and if it is ignored the deadline turns the hang into a plain failure with a message saying what happened.

Closes #8533 (closed)

🤖 Generated with Claude Code