skills: promote e2e-locators-that-lie (four Playwright harness defects on Mendix pages)

[MendixMau]

Direct lane (skills/)

What changed and why (one paragraph):
Promotes skills/e2e-locators-that-lie.md from a personal toolkit. Four Playwright harness defects, each of which produced a confident, specific verdict about the app (three false FAILs, one false PASS): (1) a Studio Pro widget name is unique per page, not per rendered DOM — a tabbed page renders hidden panes at zero height, so .first() on .mx-name-x asserts or drives the hidden node; use :visible, and for assertions check every match until one is visible; (2) click({ force: true }) on a Mendix tab is swallowed by the <li> and never throws, so the idiom force.catch(() => plain) is dead code — plain click first, force as fallback; (3) a grid re-renders on the server response, not on the click, so a single text read at settle time races — poll until the property holds within a bounded deadline, both for textPresent and textAbsent; (4) a fixture literal that the reset script itself writes measures only the reset — derive values from the rows and assert the claimed property (e.g. "nobody in both association sets") instead of a magic number.

Field evidence: measured 2026-09-17 on a user-group administration module (tabbed detail page, members grid, per-row role dropdown). The fixes exist in that project's runner (anyVisible deadline loop, :visible widget lookup, plain-then-force click, polled awaitText) and the fixture check was rewritten from MemberCount = 6 to a cross-association join. Only the patterns are quoted, no project data.

Wired: routing row in bin/lib/skill-routing.tsv (test,review · stages 5,6 · ondemand · diagnose) and re-rendered surfaces (README, ROUTING.md, test/review agent templates, gate-check stage map); one-line pointer in skills/e2e-harness-base.md "Mendix-specific quirks"; CHANGELOG line in the same commit. Cross-references point at toolkit skills only (e2e-harness-base.md, journey-proof.md, fixture-seeding.md, tool-output-is-not-ground-truth.md, and unhappy-path-testing.md, which is being promoted in a parallel PR).

Runner status (not changed here): project-tests/e2e/journey-runner.js does not apply rules 1–3 — w() is .mx-name-<n> + .first() without :visible, the click is force.catch(() => plain) (preceded by a hit-test, which catches a cover but not a handler on a child node), landing guard / widget checks use .first().isVisible(), and textPresent/textAbsent read the body once. The skill says so explicitly; porting the fixes to the runner is a follow-up instrument change with its own field proof.

Checks run (all clean): bin/check-pr-discipline.sh, bin/check-skill-routing.sh, bin/render-routing.sh --check ("surfaces in sync; all skills routed or exempted; baseline 75969 words / budget 80000"), bin/check-no-client-data.sh with the local denylist and LEAKGUARD_BASE=origin/master ("no new capitalised words; no client data in 743 tracked files"), bin/check-portability.sh ("clean, 187 files").

• No client data anywhere in the diff
• Size cap: 9 files, well under 400 lines excluding CHANGELOG.md
• Test tier reached: T0 (mechanical gates above); the skill is prose plus quoted patterns, no instrument changed
• For a new skill: routing row added (bin/lib/skill-routing.tsv + bin/render-routing.sh)
• CHANGELOG.md line appended in this PR, crediting the source
• N/A: no instrument, no bug entries

🤖 Generated with Claude Code

[MendixMau]

Merged into master via integration PR #258 (squash 9dd9be4), on the maintainer's approval. Closing this PR; its branch is part of that squash.

---

Generated by Claude Code