fix(test): keep test-local.sh from silently dying on missing ~/.lucli/express

[bpamiri]

Summary

• tools/test-local.sh silently aborts with EXIT=1 (no /tmp/wheels-test-server.log produced) on every install whose ~/.lucli/express/ directory never existed — i.e. every install since the lucli→wheels rebrand window closed.
• Root cause is the interaction between set -euo pipefail (line 26) and find ~/.wheels/express ~/.lucli/express -path "*/lib/ext" -type d 2>/dev/null | head -1 (line 81). find exits non-zero whenever any path argument is missing; 2>/dev/null hides the stderr but not the exit status. pipefail propagates it through head -1, the assignment trips set -e, the cleanup trap fires with nothing to clean up, and the user sees the script just stop after "Starting Wheels CLI server on port 8080…".
• Drop the now-dead ~/.lucli/express fallback (rename landed in 3.0; recent CLI releases extract Lucee Express to ~/.wheels/express/ only) and add || true for defense in depth so a truly fresh install (before wheels start has ever run) leaves LUCEE_LIB empty and the downstream [ -n "$LUCEE_LIB" ] guard skips the JDBC pre-install cleanly.

Test plan

• Reproduced the silent fail locally on a system with only ~/.wheels/express/ and no ~/.lucli/express/ (post-rename setup): script exits 1 after "Starting Wheels CLI server on port 8080…" with no /tmp/wheels-test-server.log.
• Confirmed find /missing /existing 2>/dev/null | head -1 exits 1 under set -euo pipefail, killing the assignment via set -e.
• After fix: bash tools/test-local.sh wheels.tests.specs.wheelstest runs end-to-end — server starts, /tmp/wheels-test-server.log produced (~69 KiB), 137 specs across 38 suites pass in ~17s.
• CI green across the compat matrix.

Follow-up (not in this PR)

While verifying, noticed a separate pre-existing bug: even after the tests pass, the script exits 1 with Tests returned HTTP 000000 because the --write-out "%{http_code}" capture (lines 129–132) appears to come back empty (likely a curl-output / response-handling quirk against the long-running test endpoint). This was previously masked by the silent-fail blocking the script from ever reaching the HTTP check. Out of scope here; will look at separately.

🤖 Generated with Claude Code

[wheels-bot]

Wheels Bot — Reviewer A

TL;DR: This PR correctly fixes a silent set -e death in tools/test-local.sh that has been breaking local test runs for every developer who installed the CLI after the lucli → wheels rebrand (i.e., anyone whose ~/.lucli/express/ never existed). The root-cause analysis is accurate, the fix is minimal and technically sound, the CHANGELOG entry is thorough, and the commit carries a DCO sign-off. Verdict: approve.

---

Correctness

The fix is correct. The failure mode is real and the shell semantics are accurately diagnosed:

• Under set -euo pipefail, find ~/.wheels/express ~/.lucli/express ... 2>/dev/null | head -1 assigns exit status 1 to the pipeline (via pipefail) when ~/.lucli/express is absent — 2>/dev/null hides stderr but not the exit code.
• The || true inside the command substitution is the right idiom: bash's set -e does not exit when a command is the left-hand side of ||, so the failing pipeline is suppressed before || true runs, and the substitution exits 0 with empty output.
• A missing ~/.wheels/express (truly fresh install, before wheels start has run) also works: find exits 1, || true fires, LUCEE_LIB="", and the downstream [ -n "$LUCEE_LIB" ] guard cleanly skips the JDBC pre-install. No silent abort.
• The removal of ~/.lucli/express as a search path is correct; per the PR body and CLAUDE.md, the rename landed in Wheels 3.0 and recent CLI releases only write to ~/.wheels/express/.

The follow-up bug mentioned in the PR body (HTTP 000 on the curl result capture) is pre-existing and appropriately out of scope.

Commits

Single commit fix(test): keep test-local.sh from silently dying on missing ~/.lucli/express (74 chars):

• Type fix, scope test — both valid per commitlint.config.js.
• Subject is under 100 chars, not ALL-CAPS.
• Commit body explains the why (not just the what), including the exact set -euo pipefail + find + pipefail chain.
• DCO sign-off present: Signed-off-by: Peter Amiri <peter@alurium.com>.

Docs

CHANGELOG [Unreleased] entry present and detailed — covers root cause, observable symptom, and the defense-in-depth rationale. No .ai/wheels/ doc updates needed (this is bash tooling, not framework behavior).

---

Nits (non-blocking):

• The PR carries the docs label but this is a tooling/infrastructure fix. Purely cosmetic; worth switching to a tooling or fix label if one exists, but doesn't affect merge.
• tools/test-local.sh:21 (pre-existing, not introduced here) still reads run \wheels browser:install`— CLAUDE.md documents the canonical form aswheels browser setup`. Out of scope for this PR; just flagging for a follow-up.

[wheels-bot]

Wheels Bot — Reviewer B (round 1)

A's review checks out. The technical analysis is accurate, the approval is backed by concrete evidence, and I found no sycophancy, false positives, or missed issues on a re-scan of the diff.

Sycophancy

None detected. A cites the exact shell semantics (find exit status surviving pipefail through head -1, tripping set -e on the assignment) rather than issuing a blanket LGTM.

False positives

None detected. I verified each claim against the diff:

• The set -euo pipefail + find + pipefail failure chain is correctly described.
• The || true idiom inside the command substitution is correctly explained: find ... | head -1 is the left-hand side of ||; when it exits non-zero (pipefail propagates find's failure even though head exits 0), bash evaluates true, the compound command exits 0, and the substitution carries empty output without triggering set -e on the assignment.
• The fresh-install path (no ~/.wheels/express/ yet) is traced correctly: find exits 1, || true fires, LUCEE_LIB="", downstream [ -n "$LUCEE_LIB" ] guard skips the JDBC step.
• DCO sign-off confirmed in the commit: Signed-off-by: Peter Amiri <peter@alurium.com>.
• CHANGELOG [Unreleased] entry confirmed in the diff.
• The wheels browser:install nit (line 21 of test-local.sh) is confirmed pre-existing and not introduced by this PR.

Missed issues

None detected. This is a bash tooling fix — cross-engine CFML concerns do not apply. The single commit is clean (type fix, scope test, subject 74 chars, DCO present). No framework behavior changed; no .ai/wheels/ doc updates needed.

Verdict alignment

A's approve verdict is consistent with the findings: minimal, correct fix, clean commit, CHANGELOG updated, no outstanding blocking concerns.

Convergence

Aligned. A's analysis is accurate, the approval is warranted, and I have no remaining disputes or missed-issue findings. Joint recommendation: approve and merge.