[saingge]

Branch: contrib/inbox-field-items, one commit (fc054c8) on origin/master 393d253.

PR body (Inbox lane: contrib/inbox/ files only)

Three items, one idea each, from a Mendix 10.24 app on mxcli v0.27.0. Each claim was re-run on 2026-10-21 on a synthetic module or on the toolkit's own fixture model, so nothing depends on the field app.

1. contrib/inbox/2026-10-21-testable-retry-loop-shape.md (582 words, skill-draft). "Retry an external call up to N times" as a loop with no return inside: lints clean (a while true with return inside gives MDL-FLOW01 once per return), mx check 0 errors, and its decision flow tests without the call. Master has nothing on retry loops (git grep -i retry -- skills contrib checked; the nearest section is "Per-row isolation in a loop" in learned-microflow-patterns.md).
2. contrib/inbox/2026-10-21-mxpatch-three-lookups-that-miss.md (519 words, learning). An entity's $Type is DomainModels$EntityImpl; a domain-model unit has no Name (so find_unit and the CLI's get/set can't select it); a flow's own Documentation is on the unit root, which find_element skips (project-bin/mxpatch.py:181). Master has EntityImpl only in an archived bug log.
3. contrib/inbox/2026-10-21-macos-mendix10-mxbuild-docker-shim.md (587 words, learning). On macOS with Mendix 10 there is no native mx/mxbuild, so the gate needs a Docker shim at MXBUILD_PATH. Gives the shim's contract in generic terms, and that bin/doctor.sh's probe_runs (lines 145-166) reads a shim that exits 3 on --version/--help as "cannot run". Master says "Docker is the only route" (skills/mpr-corruption-and-sp-load-errors.md:61) but has no contract and no probe note. Relates to Local lane: no single 'what runs where' answer (MxBuild/runtime/DB/Playwright, Docker or not); doctor-triage says the ~/.mxcli/mxbuild cache is Linux-only but it holds a working Mach-O arm64 build on macOS #525, which is about doctor-triage's "Linux only" row: item 3 adds that this is true for Mendix 10 on a Mac and (per Local lane: no single 'what runs where' answer (MxBuild/runtime/DB/Playwright, Docker or not); doctor-triage says the ~/.mxcli/mxbuild cache is Linux-only but it holds a working Mach-O arm64 build on macOS #525 and a local file check) not for the 11 builds.

• No client data: no client/vendor/person names, codenames, internal hosts, real local paths

Verification

• bin/check-no-client-data.sh: "no client data detected in 1151 tracked files".
• LEAKGUARD_BASE=origin/master bin/check-no-client-data.sh: new capitalised words HandleOutcome MaxRetries RetryState Silicon, all generic (synthetic identifiers; "Apple Silicon").
• bin/check-pr-discipline.sh: clean. git diff --stat origin/master...HEAD | tail -1: 3 files changed, 218 insertions(+).
• Item 1, on a copy of a 10.24.18 app with a synthetic Shop module (mxcli v0.27.0): mxcli check passed; exec created the flows; mxcli lint had no hit on the Shop flows; the while true-with-return variant gave 2 MDL-FLOW01 hits; mx check through the Docker shim: "The app contains: 0 errors." (52 s, exit 0, both versions present). mx-test-docker-style run of 5 tests: "5 passed, 0 failed, 0 skipped", exit 0. The same tests with one expectation flipped: "FAIL h2 ... expected $State/Attempt = 2, actual: 1", "4 passed, 1 failed", exit 1. The CE0104 statement is from check passes a statement that can never run (after return, or after a while true with no break), and mx check then fails with CE0104 / CE0105 / CE0709 mendixlabs/mxcli#1491 (read with gh issue view), not re-run. The "68 of 68" in the front matter is the field app's own count.
• Item 2: fixture model rebuilt with executescript from tests/wave2/fixtures/mxpatch/model/TestApp.mpr.sql plus mprcontents, python3 -I. find_element(doc, "Car", "DomainModels$Entity") raised NotFound; with DomainModels$EntityImpl it found Car, VCar and ApprovalContext. The domain-model unit's name is None; find_unit("DomainModels$DomainModel", "MyFirstModule") raised NotFound; mxpatch.py find printed MyFirstModule.None; get --type DomainModels$DomainModel --key Documentation printed "mxpatch: --type and --name are required". Selecting by module_of then set_value(car, "Documentation", ...) gave changed and a dry-run line. find_element(u.doc, "<microflow name>") raised NotFound for each microflow; mxpatch.py set ... --key Documentation with no --element printed "Documentation: '' -> 'Does X.'" (dry run). mxpatch.py verify: "0 mismatch(es), 0 missing file(s), 0 orphan file(s)." Not applied to a model; not opened in Studio Pro.
• Item 3: file on this Mac: the 10.24.18 cached mxbuild is "ELF 64-bit LSB pie executable, ARM aarch64"; the Studio Pro 10.24.18 bundle's Contents/modeler has 502 files and none named mx or mxbuild; the Studio Pro 11.12.4 bundle has mx and mxbuild as "Mach-O 64-bit executable arm64". The shim run is under item 1. probe_runs was extracted from bin/doctor.sh and run on two stubs: a stub that exits 3 without an .mpr returned exit 3; one that answers --version with exit 0 returned exit 0 (how=--version). The mxcli test --local darwin error text and the 110-of-110 field test run come from the project's bug log, not re-run today. Not run: the toolkit's gate on Mendix 11 on a Mac.

[MendixMau]

Thanks @saingge! Independent review: MERGE. We moved your commit unchanged, with your authorship, into #558, so CI runs without fork approval and we can handle CHANGELOG conflicts. This PR closes when #558 merges; the merge commit carries a Co-authored-by line for you.

---

Generated by Claude Code