refactor(app-restart, test-scratch): share the mxtest guard and mxcli-run pid check via _common.sh

[MendixMau]

Direct lane

What changed and why (one paragraph):
mxtest_after_startup, mxtest_left and mxtest_refusal were copied into both project-bin/app-restart.sh and project-bin/test-scratch.sh. They now live once, in a new section at the end of project-bin/_common.sh. The backlog said the two copies differed by the "Otherwise restore it" echo. In fact both copies print that line. The real difference is app-restart's else branch (the "stop the live app with Ctrl+C" advice), which test-scratch never reaches because it always passes test-scratch. The existing tool-name argument already selects that branch, so no new parameter was needed. The shared function's refusal text is byte-identical to each old copy: I checked it for both callers with HEAD set to ?, empty and a microflow name. pid_is_mxcli_run (app-restart) becomes mxtk_pid_is_mxcli_run, built on mxtk_pid_args:

• Linux with /proc: run must be a whole argument after one that names mxcli. An argument that merely contains " run " no longer counts, which is the one intended tightening.
• macOS, and Linux without /proc: the joined ps line is matched as before.
• Windows: the tasklist image-name check is kept, so there is no PowerShell call per check.

Both scripts now stop with exit 2 and the sync-project.sh --upgrade-bin _common.sh remedy when the installed _common.sh predates these helpers. This follows the existing mxtk_port_holder guard pattern. Each fixture gets one new case for this guard. The test-scratch case fails 3/3 against master's script, which shows it tests the guard. bin/lib/install-manifest.sh and sync need no change: both files are already in MXTK_PROJECT_BIN, an unedited _common.sh is refreshed through past-stock, and a hand-edited one holds back its callers (BIN_HOLD). The guard covers what is left.

Item 6 (mxtest/pid half) of the 2026-10-14 engineering backlog review. The lock half (holder_live/take_over) is a separate PR.

Field evidence — none yet. This is a behaviour-preserving refactor of existing instruments. The fixture outputs on the existing cases are identical to master's (a line-by-line diff of the ok/FAIL lines, with pids and ports normalised):

• test-app-restart: 64/0 on master → 69/0 (64 identical + 5 new)

• test-test-scratch: 46/0 → 51/0 (46 identical + 5 new)

• test-shared-lock-takeover: 6/0 → 6/0

• test-stack-up-local: 60/0 → 60/0

• test-pipefail-grep-q 3/0 and test-common-windows 10/0 also pass; both scan or source _common.sh.

• No client data anywhere in the diff

• Size cap: 6 files, ~350 lines changed, most of it a move

• Test tier reached: T1. Fixtures above, plus a Linux probe of mxtk_pid_is_mxcli_run against real processes (mxcli run --local → 0, mxcli "do run x" → 1, mxcli test → 1; macOS path forced: joined-line behaviour unchanged). No macOS or Git Bash run.

• For a changed instrument: no new parser, so no golden capture needed; layouts unaffected (no path logic changed); Windows branch kept verbatim

• For a new skill: n/a

• CHANGELOG.md line appended, crediting the source

• New bug entries: n/a

Checks: render-routing --check in sync, baseline 75,028 words before and after (budget 80,000); check-scripts 110/110; check-portability clean; leak guard clean; check-no-private-citations clean; check-pr-discipline clean.

🤖 Generated with Claude Code

https://claude.ai/code/session_01VJgWP5vEoAsNsJYqCDMGNw

---

Generated by Claude Code

[MendixMau]

Landed on master in #385 (squash 868b3cb), after its own independent review on the night integration branch. Closing this PR.

---

Generated by Claude Code