[MendixMau]

Direct lane (changes to skills/, bug-logs/, bin/, project-bin/, project-tests/, pipelines)

What changed and why (one paragraph): A program that drives project-bin/exec.sh had to infer the outcome from exit codes whose meaning differs per script; an outside review of v2026.11.12 reported an orchestrator that read exit 4 ("applied, gate passed, lint only") as a failure. exec.sh now prints one fixed line on every path that reaches a verdict: MXTK-RESULT exec verdict=<applied|applied-lint|applied-unverified|partial|restored|failed|refused> applied=<yes|no|partial|unknown> verified=<yes|no|skipped> lint_new=<N|unknown|blind|skipped|na> exit=<code>. It is built in one function and printed before each path's closing human lines, so it is never the last line. Purely additive: no exit code changes and no existing output line is changed or removed. Paths that never reach a verdict (usage errors, a missing .mpr, an abort under set -e) print no line, and the header says that no line means no verdict, never success. Restore paths that had no closing line of their own get one; it says "rolled back", "NOT rolled back" or "no rollback was attempted" according to what happened. A short pointer is added to skills/harness-architecture.md (on demand).

Impact

Row	Answer
Process — human steps or gates added/removed; does anything new block; new stage artifacts	none.
Dependencies — shared files touched; order vs other open PRs; new tools/platform assumptions; new artifact reads + their producer	project-bin/exec.sh (additions only; four existing lines gain a trailing ; LINT_NEW=…), a new fixture, skills/harness-architecture.md, CHANGELOG.md. A later PR will rewrite the orchestrator contract section of the same skill on top of this one.
Context cost	none: baseline 66,025 → 66,025 of 80,000; the touched skill is on demand.
Behaviour	n/a: scripts only (the skill change is a pointer).
Runtime	Measured on a real project, median of 3 on Linux: 20.5 s → 20.9 s for an applied script, 0.57 s → 0.49 s for a refused one: within run-to-run variation. Time cost: per exec · 0 s measured on a cloud Linux container · can force rework: n · offset: none needed

Field evidence — a fictional test app built with the toolkit on 2026-11-12 (Mendix 11.13, mxcli v0.28), real mxbuild gate. Old and new exec.sh on the same script: same exit code, and the output differs by exactly the one added line. Applied: MXTK-RESULT exec verdict=applied applied=yes verified=yes lint_new=blind exit=0. Refused on purpose (reference to an entity that does not exist): MXTK-RESULT exec verdict=refused applied=no verified=skipped lint_new=na exit=1. Fixtures: test-exec-result-line.sh 72 of 72 (stub-based; the line carries no tool output), and the existing exec fixtures unchanged: gate-retry 52, lint-diff 9, lint-autorestore 17, crash-rollback 13, hash-verify 49, gate-kind 32, lock-watch 104, all passing.

Independent review: MERGE-AFTER-FIX (verdicts failed and partial and the --patch paths untested; two paths disagreed when a restore itself fails; lint_new=na covered three different cases), fix round, MERGE-AFTER-FIX again (the new closing line said "restore" on two paths that never attempt one), fixed, then MERGE, no findings at 29ca2b78. Not run on macOS or Git Bash.

• No client data anywhere in the diff
• Size cap: 1 script (+about 60 lines), 1 new fixture, 1 skill pointer, CHANGELOG.md
• Test tier reached: fixtures plus a before/after run on a real project with the real build gate
• For a new/changed instrument: expected lines pasted from actual runs; layouts unaffected (no path logic touched); macOS and Git Bash considered, not run
• For a new skill: n/a
• CHANGELOG.md line appended in this PR, crediting the source
• New bug entries: n/a

---

Generated by Claude Code