Fix shell lint baseline warnings

[arittr]

What problem are you trying to solve?

After #1676 introduced scripts/lint-shell.sh --all, the first warning-level ShellCheck baseline run failed on existing tracked shell scripts. That meant the repo would have a useful shell lint command, but agents and reviewers would not be able to treat new warning-level findings as regressions because the baseline already started red.

The initial baseline exposed 14 warning-level findings: SC2155 in tests/claude-code/test-helpers.sh, SC2088 in tests/claude-code/test-worktree-path-policy.sh, SC2034 in the brainstorming server start/stop scripts, SC2064 and SC2320 in the subagent integration test, and SC2164 in skills/brainstorming/scripts/start-server.sh.

What does this PR change?

Fixes the warning-level ShellCheck findings exposed by #1676 without changing the tooling branch. The fixes split declaration/assignment pairs, preserve the real pipeline exit status in the integration test, keep legacy ~/.config/superpowers/worktrees assertions literal, handle cd failure in start-server.sh, and rename intentionally-unused loop variables.

Is this change appropriate for the core library?

Yes. These are small correctness and maintainability fixes to existing core shell scripts/tests so the repository can adopt a clean shell lint baseline. This does not add a dependency, does not add a new harness, does not introduce project-specific configuration, and does not modify behavior-shaping skill prose.

What alternatives did you consider?

1. Suppress the warnings inline with ShellCheck directives: rejected because these were straightforward to fix directly, and suppressions would teach agents to paper over useful findings.
2. Lower the baseline below warning severity: rejected because warning severity is the useful minimum for catching likely shell hazards.
3. Fold these fixes into Add shell lint script #1676: rejected to keep the tooling PR reviewable. Add shell lint script #1676 introduces the script; this stacked PR makes the existing repo pass the new baseline.
4. Focus on shfmt first: rejected because formatter churn is separate from warning-level correctness. shfmt can be a later mechanical PR.

Does this PR contain multiple unrelated changes?

No. Every file change is directly tied to making the ShellCheck warning baseline pass. The branch is stacked above #1676 because it depends on the lint script introduced there.

Existing PRs

• I have reviewed all open AND closed PRs for duplicates or prior art
• Related PRs:
  • Add shell lint script #1676: direct dependency; introduces scripts/lint-shell.sh and the test suite for the tooling.
  • feat: add Antigravity CLI (agy) support #1657: motivating Antigravity harness PR that added/changed Bash and prompted the maintainer discussion about better Bash guardrails.
  • feat(tests): cross-harness portability lint #1603: prior static-lint infrastructure PR for cross-harness portability; related as repo-maintenance linting prior art, but it does not address shell scripts.

Searches reviewed: shell lint, lint-shell, ShellCheck, shfmt, and bash lint across open and closed PRs. I did not find an earlier PR fixing a repository-wide ShellCheck baseline for tracked shell scripts.

Environment tested

Harness (e.g. Claude Code, Cursor)	Harness version	Model	Model version/ID
Codex desktop + local shell	Not exposed in session	GPT-5 Codex	Not exposed in session

Local verification performed on macOS / Darwin 25 with Bash, ShellCheck, and shfmt available.

Commands run:

git diff --check
scripts/lint-shell.sh --all
bash tests/claude-code/test-worktree-path-policy.sh
bash tests/shell-lint/test-lint-shell.sh
bash tests/brainstorm-server/windows-lifecycle.test.sh

Observed results: scripts/lint-shell.sh --all passes on the rebased stack, linting 31 shell files. The shell lint tests and worktree path policy test passed. The brainstorming server lifecycle test passed with 9 passed, 0 failed, and 3 Windows-specific skips on macOS.

New harness support (required if this PR adds a new harness)

Not applicable. This PR does not add a new harness.

Evaluation

• Initial prompt: Drew asked to run the new shell baseline, then update tooling first and start fixing the scripts themselves on a new stacked branch above the tooling PR.
• How many eval sessions after making the change: 0 agent-behavior eval sessions. This is a shell-script cleanup, not a behavior-shaping skill change.
• Observed outcome: Before this PR, scripts/lint-shell.sh --all failed with 14 warning-level ShellCheck findings. After this PR, the same baseline passes on the rebased stack. The targeted tests around the changed helper/policy/lifecycle paths also pass.

Rigor

• N/A: this is not a skills change, so superpowers:writing-skills adversarial testing is not required.
• This change was tested against the failing baseline, not only by inspection: the warning-level ShellCheck baseline was run before fixing and again after fixing, and targeted tests were run for changed shell helper/policy/lifecycle behavior.
• I did not modify carefully-tuned content (Red Flags table, rationalization lists, "human partner" language).

Human review

• Drew Ritter, a maintainer, directed this work in-session, reviewed the scope and tradeoffs, and requested the commit/stack. This PR body update exists to document that human involvement explicitly.

AI was used for implementation assistance.

_{Stack created with GitHub Stacks CLI • Give Feedback}