diff --git a/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-19-brain-list-questions-among-decisions.json b/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-19-brain-list-questions-among-decisions.json new file mode 100644 index 0000000000..dd46df6eef --- /dev/null +++ b/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-19-brain-list-questions-among-decisions.json @@ -0,0 +1 @@ +{"area": "cmd/mxcli/brain", "date": "2026-11-19", "symptom": "`brain list --json` reports open questions with kind \"decision\" (and `--kind decision` returns them), so a tool counting decisions counts every open question twice (ako/PizzaV1's overview had to filter `open: true` out by hand)", "cause": "List set each entry's kind with EntryKind(), which reads an open question as a decision; the stored form has no question kind", "file": "`cmd/mxcli/brain/list.go` (Store.List)", "insight": "Give the listing a reporting-only kind (KindQuestion) for open entries rather than a stored kind: EntryKind() keeps reading them as decisions for every internal consumer, and the JSON and --kind filter stop conflating them", "refs": ["ako/mxcli#2443"], "test": "TestListFiltersTheCommittedStore"} diff --git a/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-19-failed-test-run-drops-reports.json b/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-19-failed-test-run-drops-reports.json new file mode 100644 index 0000000000..b92d5cca12 --- /dev/null +++ b/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-19-failed-test-run-drops-reports.json @@ -0,0 +1 @@ +{"area": "cmd/mxcli", "date": "2026-11-19", "symptom": "report all --run in an environment where the tests cannot run: tests AND traceability 'Not generated: open out/junit.xml: no such file or directory', though --journeys/--spans were given", "cause": "collectEvidence recorded the failed `mxcli test` as a warning but then failed the whole report reading the results file the run never wrote", "file": "`cmd/mxcli/cmd_report_evidence.go` (collectEvidence)", "insight": "Evidence a report can stand without is a warning when it is missing, not an error: the error hid the real cause (the failed run) behind a file-not-found, and took the journeys down with it", "refs": ["ako/mxcli#2452"], "test": "TestMissingJUnitIsAWarningNotAFailure"} diff --git a/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-19-perf-step-counts-background-spans.json b/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-19-perf-step-counts-background-spans.json new file mode 100644 index 0000000000..b6b0d57cec --- /dev/null +++ b/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-19-perf-step-counts-background-spans.json @@ -0,0 +1 @@ +{"area": "cmd/mxcli", "date": "2026-11-19", "symptom": "`report performance` shows random steps several seconds slower than they are: on ako/PizzaV1 'place order and pay' read 4,884 ms of server time while its own requests took 280 ms", "cause": "spansIn gave a step every span that started inside its time window, so a queued task (the simulator's 1.5 s clock tick), scheduled events and the task queue's polling counted as the step's server time, database time and queries", "file": "`cmd/mxcli/projectreport/performance.go` (spanIndex.stepSpans)", "insight": "Attribute by trace, not by time: a step owns the traces rooted in an HTTP request to the app port (the most common server.port among http roots) that start in its window; other roots are background, measured apart. With that, db time matched the project's own analysis on all 12 steps. Keep the window rule only as a labelled fallback for collectors without http attributes", "refs": ["ako/mxcli#2443"], "test": "TestPerformanceCountsOnlyTheStepsOwnRequests"} diff --git a/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-19-quality-html-score-bars-empty.json b/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-19-quality-html-score-bars-empty.json new file mode 100644 index 0000000000..c988aed5a3 --- /dev/null +++ b/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-19-quality-html-score-bars-empty.json @@ -0,0 +1 @@ +{"area": "cmd/mxcli", "date": "2026-11-19", "symptom": "`mxcli report --format html` draws every score bar as an empty grey track, whatever the score (ako/PizzaV1 shipped tools/fix_quality_html.py to patch it)", "cause": "The linter's own HTML formatter gave .score-fill a width but no display rule, so the inline span ignored its width; the scored report also had a style of its own, apart from every report kind", "file": "`cmd/mxcli/projectreport/quality.go` (QualityDoc) and `doc.go` (.meter)", "insight": "Render the scored report's page through the shared Doc renderer instead of fixing the one-off CSS: one style to maintain, and the bar is a block element with a coloured fill. The renamed class (.meter) also avoids the software map's own .bar control rows, which a shared .bar rule squashed", "refs": ["ako/mxcli#2443"], "test": "TestQualityDocBarsHaveAFill"} diff --git a/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-19-report-all-inputs-to-wrong-reports.json b/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-19-report-all-inputs-to-wrong-reports.json new file mode 100644 index 0000000000..8713c5a496 --- /dev/null +++ b/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-19-report-all-inputs-to-wrong-reports.json @@ -0,0 +1 @@ +{"area": "cmd/mxcli", "date": "2026-11-19", "symptom": "report all: traceability proves 5 of 111 requirements where report traceability proves 101, readiness can never pass 'not assessed in full' (no way to give test results), and performance shows one 'step' per functional journey, each over budget", "cause": "passThrough forwarded --journeys/--spans to performance and tests only, report all had no --junit/--run/--tests, and one journey log fed both the functional reports and performance", "file": "`cmd/mxcli/cmd_report_all.go` (passThrough, needsInput)", "insight": "A roll-up is only as right as the inputs it hands on: test it per receiving report (TestReportAllHandsEachReportItsOwnInputs), not only that every page is written. A functional log and a performance run are different evidence and get different flags; --run runs once and leaves junit.xml for the next report to read", "refs": ["ako/mxcli#2451"], "test": "TestReportAllHandsEachReportItsOwnInputs"} diff --git a/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-19-security-caller-check-own-body-only.json b/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-19-security-caller-check-own-body-only.json new file mode 100644 index 0000000000..0e07588377 --- /dev/null +++ b/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-19-security-caller-check-own-body-only.json @@ -0,0 +1 @@ +{"area": "cmd/mxcli", "date": "2026-11-19", "symptom": "the security report flags a callable flow high although its sub-microflows show it is not (Shop.DS_Info_Line on ako/PizzaV1: takes a Product, creates the caller's own cart line), and labels flows that commit the parameter's entity in a sub as reading only", "cause": "collectCallableFlows looked at the flow's own body only, and 'writes' meant any write to any entity; entity access off + any write was high whatever the parameter entity's read rule", "file": "`cmd/mxcli/projectreport/security.go` (collectCallableFlows, anyOnEntity)", "insight": "Follow the call closure, but judge it on the entities the flow takes: a sub that stamps [%CurrentUser%] on a Toast turned 19 PizzaV1 flows 'ok' in a first cut, and creating one's own cart line is not changing what the client chose. The client can only pass objects its role may read, so entity access off + a write is high only when the read rule is 'all'; otherwise the read rule bounds it (PizzaV1's hand review: 'only the guest's own rows'). Compare a new verdict rule against a hand-reviewed app before trusting it", "refs": ["ako/mxcli#2443"], "test": "TestCallableFlowFollowsSubMicroflows"} diff --git a/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-19-spans-gz-not-read.json b/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-19-spans-gz-not-read.json new file mode 100644 index 0000000000..a3abf1ec02 --- /dev/null +++ b/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-19-spans-gz-not-read.json @@ -0,0 +1 @@ +{"area": "cmd/mxcli", "date": "2026-11-19", "symptom": "--spans after-spans.jsonl.gz fails: \"line 1: invalid character '\\x1f' looking for beginning of value\"", "cause": "readSpanFile parsed the file's bytes as JSONL/OTLP without checking for gzip", "file": "`cmd/mxcli/projectreport/otlp.go` (readSpanFile)", "insight": "Decide on the gzip magic bytes (1f 8b), not the extension, then pick the format from the name without .gz — so a .pb.gz is still protobuf and a mis-named file still reads", "refs": ["ako/mxcli#2452"], "test": "TestReadSpansGzip"} diff --git a/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-19-unused-annotation-without-module.json b/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-19-unused-annotation-without-module.json new file mode 100644 index 0000000000..1bedf408a6 --- /dev/null +++ b/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-19-unused-annotation-without-module.json @@ -0,0 +1 @@ +{"area": "cmd/mxcli", "date": "2026-11-19", "symptom": "an architecture-report annotation `unused:FaultProbe` changes nothing; the document stays 'verify' and the key is listed under 'annotations that matched nothing' (ako/PizzaV1 had to rename every key to `unused:Shop.FaultProbe`)", "cause": "the page looked the key up by the qualified name only, while the unused table and the natural way to write the key both say just the document name", "file": "`cmd/mxcli/projectreport/architecture.go` (RenderArchitecture, unused table)", "insight": "Accept the short key only when the short name is unique among the unused documents; an ambiguous one stays visibly unmatched rather than annotating the wrong document", "refs": ["ako/mxcli#2443"], "test": "TestUnusedAnnotationWithoutModule"} diff --git a/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-20-gate-list-reached-projects-without-baselines.json b/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-20-gate-list-reached-projects-without-baselines.json new file mode 100644 index 0000000000..a1aef75906 --- /dev/null +++ b/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-20-gate-list-reached-projects-without-baselines.json @@ -0,0 +1 @@ +{"area": "cmd/mxcli", "date": "2026-11-20", "symptom": "Making `verify` the gate in the generated CLAUDE.md (ako/mxcli#2438) was meant to reach new projects only. It reached every bootstrapped project: the session-start bootstrap runs `mxcli init --sync-skills .` each session, which regenerates the mxcli section of CLAUDE.md, so an upgraded mxcli replaced `lint` and `docker check` with `verify` in projects that had no verify-baseline/. Without baselines verify reports every finding the project already had as new, and a project with errors fails on its whole backlog.", "cause": "The gate list was a single package-level list chosen when mxcli was built, and the generator had no way to see the project: ToolFile.Content takes only the project name and the .mpr file name. I assumed init-time generation (the PR text and proposal question 12 said existing projects keep their old list) without reading what `--sync-skills` regenerates. The sync path (init_tooling_sync.go) re-renders owned docs from the same list on every session start.", "file": "`cmd/mxcli/init_claudemd.go` (classicGates, verifyGates, hasVerifyBaseline, projectGatesFor, generateClaudeMDIn), `cmd/mxcli/tool_templates.go` (ToolFile.ContentIn, render), `cmd/mxcli/init.go`, `init_tooling_sync.go`, `cmd_add_tool.go`; tests in `cmd/mxcli/init_claudemd_gates_test.go`.", "insight": "Before changing generated text that lives in a project's own files, find every path that regenerates it, not only the one that creates it: a sync that runs unattended every session turns a binary upgrade into a change of every project's instructions. The fix makes the generated list a function of the project (the verify gate list only when verify-baseline/lint.json exists), so recording the baselines moves a project onto verify at the next sync and an upgrade alone moves nothing. A test must exercise the sync path with and without the file (TestSyncMovesAProjectOnAndOffTheVerifyGateList), not only the generator, because the generator alone cannot show who calls it. Found by asking what the risk of the default was and then reading init --sync-skills; the ChipCoV14 clone's own .claude/bootstrap-mxcli.sh showed the call."} diff --git a/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-20-verify-apply-failed-later-stage-says-applied.json b/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-20-verify-apply-failed-later-stage-says-applied.json new file mode 100644 index 0000000000..bf72434431 --- /dev/null +++ b/.claude/skills/fix-issue/findings/cmd-mxcli/2026-11-20-verify-apply-failed-later-stage-says-applied.json @@ -0,0 +1 @@ +{"area": "cmd/mxcli", "date": "2026-11-20", "refs": ["ako/mxcli#2468", "ako/mxcli#2447"], "ce": [], "symptom": "`mxcli verify --apply` that passed check and exec and then failed mxcheck printed `Result: FAILED (exit 20)` exactly like a gate that wrote nothing, although exec had already written the project; the broken change stayed in the project until someone noticed.", "file": "cmd/mxcli/verify/render.go WriteText; test render_test.go TestWriteTextSaysTheScriptsWereAppliedWhenALaterStageFails; help text in cmd/mxcli/cmd_verify.go", "cause": "The text renderer printed the same Result line for every non-zero exit; Result.Applied (exec ran and wrote) was in the JSON but never read by the text output.", "insight": "**The signal already existed; the renderer ignored it.** Result.Applied was set when exec applied the scripts, so the fix reads it rather than adding state. The test needs three controls — a passing run, a run that never applied, and a check that failed before exec — because the sentence is only true when exec wrote AND something later failed; a stub of the condition fails exactly one test, the positive one."} diff --git a/.claude/skills/fix-issue/findings/mdl-backend/2026-11-19-rest-base-url-constant-dropped.json b/.claude/skills/fix-issue/findings/mdl-backend/2026-11-19-rest-base-url-constant-dropped.json new file mode 100644 index 0000000000..5c9844517c --- /dev/null +++ b/.claude/skills/fix-issue/findings/mdl-backend/2026-11-19-rest-base-url-constant-dropped.json @@ -0,0 +1 @@ +{"area": "mdl/backend", "date": "2026-11-19", "symptom": "a consumed REST service whose base URL is a constant shows an empty BaseUrl in the catalog (CATALOG.CONSUMED_REST_SERVICES), indistinguishable from one with no base URL", "cause": "the reader type-asserted Rest$ValueTemplate only; a Rest$ConstantValue base URL fell through and was dropped", "file": "`mdl/backend/modelsdk/integration_read.go` (restBaseURL)", "insight": "A gen property typed element.Element is a union: read every concrete type it can hold, not the one the writer emits. Kept apart from BaseUrl (BaseUrlConstant) so describe and its round trip, which only know literals, are unchanged", "refs": ["ako/mxcli#2426"], "test": "TestRestBaseURLReadsTheConstant"} diff --git a/.claude/skills/fix-issue/findings/mdl-backend/2026-11-20-list-markers-js-call-module-security.json b/.claude/skills/fix-issue/findings/mdl-backend/2026-11-20-list-markers-js-call-module-security.json new file mode 100644 index 0000000000..5bf1819bc0 --- /dev/null +++ b/.claude/skills/fix-issue/findings/mdl-backend/2026-11-20-list-markers-js-call-module-security.json @@ -0,0 +1 @@ +{"area": "mdl-backend", "date": "2026-11-20", "refs": ["ako/mxcli#2467", "ako/mxcli#258"], "ce": [], "symptom": "Studio Pro rewrote mxcli-made units on first open: an empty Security$ModuleSecurity.ModuleRoles stored with marker 1 and a Microflows$JavaScriptActionCallAction.ParameterMappings stored with marker 3, where Studio Pro writes 3 and 2. check, exec and mx check passed; only the git diff after opening showed it.", "file": "mdl/backend/modelsdk/module_write.go (ModuleSecurity TypeDefaults), mdl/backend/modelsdk/microflow_write.go (JavaScriptActionParameterMapping). Test: mdl/backend/modelsdk/list_markers_2467_test.go", "cause": "ModuleRoles was registered as marker 1 on the reading that it is a by-name role list; it holds the roles themselves. The JS mapping child type was never registered, so it fell back to the encoder default 3.", "insight": "**A list's marker follows what it holds, not what it is called**: ModuleRoles is 1 on UserRole (names) and 3 on ModuleSecurity (parts). Settle a marker with a census of a Studio Pro-made project (owner $Type.key -> marker counts), never by analogy. Register it per owner/key too, so canon.NormalizeListMarkers corrects the units an older mxcli already stored on their next write."} diff --git a/.claude/skills/fix-issue/findings/mdl-backend/2026-11-20-progress-bar-null-message.json b/.claude/skills/fix-issue/findings/mdl-backend/2026-11-20-progress-bar-null-message.json new file mode 100644 index 0000000000..0427fb4561 --- /dev/null +++ b/.claude/skills/fix-issue/findings/mdl-backend/2026-11-20-progress-bar-null-message.json @@ -0,0 +1 @@ +{"area": "mdl-backend", "date": "2026-11-20", "refs": ["ako/mxcli#2448", "mendixlabs/mxcli#1902"], "ce": [], "symptom": "`call microflow … with (ProgressBar: Blocking)` without ProgressMessage stored Forms$MicroflowSettings.ProgressMessage as null; check, exec and mx check passed, the project opened, and Studio Pro 10.24.28 crashed exporting the page (NullReferenceException in MicroflowSettings.WriteProperties) and errored on typing in the progress-message field.", "file": "mdl/backend/modelsdk/widget_write.go progressMessageFor (microflowSettingsToGen, CallNanoflowClientAction). Test: mdl/backend/modelsdk/progress_message_null_2448_test.go", "cause": "The writer set ProgressMessage only when the script gave one; RegisterTypeDefaults lists it under NullFields, which is right only for ProgressBar None.", "insight": "**Null is a valid default only for the value of the enum it belongs to.** The reporter's census of a real 10.24 project (None+null 210, Blocking+Text 22, NonBlocking+Text 11, Blocking+null 2 — both mxcli's, both crashing) is the evidence; MxBuild 11.12.6 --target=deploy built both shapes without error, so the crash lives in Studio Pro's export, which no headless check here reaches. Decide such a default from Studio Pro's stored data, not from what the checker accepts."} diff --git a/.claude/skills/fix-issue/findings/mdl-executor/2026-11-19-drop-member-still-referenced-ce1613.json b/.claude/skills/fix-issue/findings/mdl-executor/2026-11-19-drop-member-still-referenced-ce1613.json new file mode 100644 index 0000000000..41b950c10b --- /dev/null +++ b/.claude/skills/fix-issue/findings/mdl-executor/2026-11-19-drop-member-still-referenced-ce1613.json @@ -0,0 +1 @@ +{"area": "mdl/executor", "date": "2026-11-19", "refs": ["ako/mxcli#2456", "ako/mxcli#1258"], "ce": ["CE1613", "CE0161"], "symptom": "`alter entity E drop attribute \"A\"`, `drop entity E` and `drop association X` passed `check` (with or without --references) when a microflow, nanoflow, page or snippet still used the target; `exec` applied the drop with at most a \"may still reference\" warning (`drop entity` printed nothing) and `mx check` then reported CE1613 on each user (CE0161 where the attribute was only in an XPath).", "file": "mdl/executor/validate_dropped_member_references.go CheckDroppedMemberReferences (MDL108), called from cmd/mxcli/cmd_check.go in the catalog-backed tier. Tests: validate_dropped_member_references_test.go", "cause": "CheckEntityMemberDrops (MDL087) only reports members a `create or modify` loses without being asked; an explicit drop is excluded by intent, and nothing asked who else uses the dropped member. The catalog already holds the answer (`show references to`).", "insight": "**The reference graph is the whole rule; the work is deciding what NOT to report.** Measured one project per case on MxBuild 11.16.1: microflow, nanoflow, page and snippet users are CE1613, an XPath-only user is CE0161, dropping an entity that only an association points at is clean (the association goes with it), and dropping then re-adding the attribute is clean. So the rule reads only those four source types, ignores documents the script itself drops, creates or alters (it may fix the reference), skips a member the script puts back, and opens the FULL catalog (references are not in fast mode) only when the script has such a drop. `exec` still applies the drop: the project tier is check-only, like MDL087, so refusing there is a separate decision."} diff --git a/.claude/skills/fix-issue/findings/mdl-executor/2026-11-20-change-handler-rejoins-late.json b/.claude/skills/fix-issue/findings/mdl-executor/2026-11-20-change-handler-rejoins-late.json new file mode 100644 index 0000000000..d77829c394 --- /dev/null +++ b/.claude/skills/fix-issue/findings/mdl-executor/2026-11-20-change-handler-rejoins-late.json @@ -0,0 +1 @@ +{"area": "mdl-executor", "date": "2026-11-20", "refs": ["ako/mxcli#2421", "mendixlabs/mxcli#1898"], "ce": [], "symptom": "In `change $Order … on error without rollback begin … end error; set $N = $N + 10; commit $Order on error …`, the change's handler rejoined at the commit handler's merge AFTER the commit, so at runtime an error in the change skipped the set and the commit. check, exec and mx check reported nothing; describe showed one shared `join rejoin1`.", "file": "mdl/executor/cmd_microflows_builder_actions.go addChangeObjectAction / addChangeVariableAction / addReplaceListAction pass no skip variable to finishCustomErrorHandler. Test: mdl/executor/error_handler_rejoin_change_2421_test.go", "cause": "finishCustomErrorHandler's skip variable defers a handler's rejoin until no later statement references it — right for an activity that CREATES the variable (it does not exist on the error path), but change/set/replace passed the variable they MODIFY, which exists on both paths, so every later use of $Order pushed the rejoin further down.", "insight": "**A 'skip' variable must be one the activity produces, not one it touches.** The skip exists so a handler does not rejoin before statements that need an object the failed activity never made; passing a pre-existing variable turns that into 'skip every later statement that mentions it'. Measured on 11.12.6: ChangeSetCommit and a three-change chain each rejoin right after their own activity; mx check 0 errors. `create` keeps the skip (control test)."} diff --git a/.claude/skills/fix-issue/findings/mdl-executor/2026-11-20-create-layout-folder.json b/.claude/skills/fix-issue/findings/mdl-executor/2026-11-20-create-layout-folder.json new file mode 100644 index 0000000000..8d3d476b84 --- /dev/null +++ b/.claude/skills/fix-issue/findings/mdl-executor/2026-11-20-create-layout-folder.json @@ -0,0 +1 @@ +{"area": "mdl-executor", "date": "2026-11-20", "refs": ["ako/mxcli#2460", "mendixlabs/mxcli#1905"], "ce": [], "symptom": "`create layout M.L (layouttype: 'Responsive', folder: 'Layouts')` passed check, and exec refused it: 'unknown property \"folder\"'.", "file": "mdl/executor/cmd_pages_layout_v3.go buildLayoutV3 (Folder via pb.resolveFolder) and the rewrite path (applyDocumentFolder when the header names a folder); mdl/executor/cmd_pages_describe.go describeLayout prints folder:. Test: mdl/executor/layout_folder_2460_test.go", "cause": "The layout header's allow-list (added to stop silently ignored properties) did not include folder, which pages and snippets take; check does not run the builder, so only exec saw the refusal; and describe never printed a layout's folder, so a round trip would have moved a foldered layout to the module root.", "insight": "**An allow-list that refuses unknown properties has to list every property its siblings accept, and check must agree with it.** Measured on 11.12.6: created in Layouts/Web, describe prints folder:, describe → exec Unchanged, a changed folder moves the layout, mx check 0 errors."} diff --git a/.claude/skills/fix-issue/findings/mdl-executor/2026-11-20-js-action-nanoflow-argument.json b/.claude/skills/fix-issue/findings/mdl-executor/2026-11-20-js-action-nanoflow-argument.json new file mode 100644 index 0000000000..cfc6fde3c7 --- /dev/null +++ b/.claude/skills/fix-issue/findings/mdl-executor/2026-11-20-js-action-nanoflow-argument.json @@ -0,0 +1 @@ +{"area": "mdl-executor", "date": "2026-11-20", "refs": ["ako/mxcli#2450", "mendixlabs/mxcli#1904"], "ce": ["CE0115", "CE0117"], "symptom": "`call javascript action … (callback = Module.Flow, other = empty)` for Nanoflow-typed parameters was written as Microflows$BasicCodeActionParameterValue with the name as expression: check and exec passed, mx check CE0115/CE0117. describe left Nanoflows$NanoflowParameterValue mappings out, so editing a Studio Pro call spliced them away (8 mappings → 3).", "file": "sdk/microflows NanoflowParameterValue; mdl/backend/modelsdk microflow_read_actions.go / microflow_write.go (Nanoflows$NanoflowParameterValue), java_write.go codeActionParamTypeToGen (NanoflowType → JavaScriptActions$NanoflowJavaScriptActionParameterType); mdl/executor addCallJavaScriptActionAction, nanoflowParamArgRefusal, describe in cmd_microflows_format_action.go. Tests: *_2450_test.go in executor and modelsdk", "cause": "The JS-action call builder knew only entity-type vs basic parameters, the code-action value model had no nanoflow variant, and describe silently skipped values it could not print. Measuring it found a second gap: create javascript action stored a Nanoflow parameter as String, so no mxcli-authored action could even have one.", "insight": "**A describer that skips an unknown value turns every later edit into data loss.** The splice rebuilds an edited activity from the description, so whatever describe leaves out is deleted on the next change while 'Unchanged' on an untouched re-run hides it. Follow the existing precedent (Java MicroflowParameterValue, #1210/#1282) for name-not-expression and the empty form. Measured on 11.12.6: NanoflowParameterValue written, mx check 0 errors (was CE0117), describe → exec Unchanged, edited call keeps both nanoflow arguments."} diff --git a/.claude/skills/fix-issue/findings/mdl-executor/2026-11-20-member-kind-and-code-action-missing-args.json b/.claude/skills/fix-issue/findings/mdl-executor/2026-11-20-member-kind-and-code-action-missing-args.json new file mode 100644 index 0000000000..dc7e1c77d6 --- /dev/null +++ b/.claude/skills/fix-issue/findings/mdl-executor/2026-11-20-member-kind-and-code-action-missing-args.json @@ -0,0 +1 @@ +{"area": "mdl-executor", "date": "2026-11-20", "refs": ["ako/mxcli#2464", "mendixlabs/mxcli#1909"], "ce": ["CE0115", "CE0117"], "symptom": "`$Guid = call java action CommunityCommons.getGUID(); change $Obj (Code = $Guid);` — a missing argument and an Integer result assigned to a String attribute — passed check and exec; mx check reported CE0115 and CE0117.", "file": "mdl/exprcheck/call_argument.go checkMemberKind (wired in parser.go); mdl/exprcheck/adapters WithCallResultKinds + flowScope.recordCallResults; mdl/executor/typecheck.go callResultKind; mdl/executor/validate.go missingCodeActionArgs. Tests: mdl/exprcheck/member_kind_2464_test.go, mdl/executor/code_action_missing_args_2464_test.go", "cause": "Three gaps: the expectations table resolves create/change members 'AttributeOf:Parent' but nothing compared value and attribute (even `Code = 5` passed); call results were never typed in the flow scope; and code-action calls were checked for unknown argument names but not for missing ones.", "insight": "**Measure which omissions the build rejects before refusing them.** A blanket 'every parameter needs an argument' rule looked right from the issue, but 11.12.6 with each kind left out showed only the entity-type parameter (ENTITY ) is CE0115 — Integer, object, type-parameter object and Microflow build clean with no mapping — so the blanket rule would have refused valid scripts. The member matrix (13 microflows, 7 CE0117) matched the #2096 call-argument predicate exactly, so it is reused rather than re-derived."} diff --git a/.claude/skills/fix-issue/findings/mdl-executor/2026-11-20-rest-call-expressions-unchecked.json b/.claude/skills/fix-issue/findings/mdl-executor/2026-11-20-rest-call-expressions-unchecked.json new file mode 100644 index 0000000000..d7fc7e0efd --- /dev/null +++ b/.claude/skills/fix-issue/findings/mdl-executor/2026-11-20-rest-call-expressions-unchecked.json @@ -0,0 +1 @@ +{"area": "mdl-executor", "date": "2026-11-20", "refs": ["ako/mxcli#2408", "mendixlabs/mxcli#1896"], "ce": ["CE0117"], "symptom": "`base64Encode(…)` (not a Mendix function) in a `call rest service` header value, and a made-up function in its body or URL, passed check and exec; mx check reported CE0117 at the Call REST service activity, while the same call in a declare was refused as MDL044.", "file": "mdl/executor/validate_microflow.go checkStmtExprFunctions (RestCallStmt: URL, URL template params, headers, auth, body, body template params, timeout). Test: mdl/executor/rest_call_expr_functions_2408_test.go", "cause": "checkStmtExprFunctions is the single list of expression sites MDL044 walks, and RestCallStmt was not in it.", "insight": "**A single list of expression sites is only as complete as its last statement type.** #1033 (nanoflows), #2199 (show message) and now the REST call were each the same omission; when a statement grows expression fields, add it to checkStmtExprFunctions in the same change."} diff --git a/.claude/skills/fix-issue/findings/mdl-other/2026-11-19-model-name-in-string-literal-rename-trap.json b/.claude/skills/fix-issue/findings/mdl-other/2026-11-19-model-name-in-string-literal-rename-trap.json new file mode 100644 index 0000000000..0a5a450586 --- /dev/null +++ b/.claude/skills/fix-issue/findings/mdl-other/2026-11-19-model-name-in-string-literal-rename-trap.json @@ -0,0 +1 @@ +{"area": "mdl/linter", "date": "2026-11-19", "symptom": "A microflow names another model element as TEXT (a Java action receives the string \"Simulation.SIM_Tick\"), so no reference exists for any checker to follow. Renaming or moving the named element passes Studio Pro, mxbuild and `mxcli check --references`, and the app fails at runtime — in the benchmark's app, the simulation stopped with an ERROR in the server log every second.", "cause": "Not a defect in mxcli: a gap in what any checker can see. Added lint rule MPR018, which scans the string LITERALS inside expression/value fields of flow actions and reports a literal that is exactly a qualified name of this project. Two design choices carry the precision: (1) literals, not whole expressions, because an expression's own references are also dotted ($Var/Mod.Assoc/Attr, Mod.Enum.Value) and those ARE real references Studio Pro renames — scanning expressions would report every association path in the project; (2) fields named Expression/Argument/Value only, found by reflection, because the sibling fields that hold references are named after what they reference (Parameter, Association, ListVariable, and JavaActionCallAction.JavaAction, MicroflowCall.Microflow) and reporting those would flag every call in the project.", "file": "`mdl/linter/rules/mpr018_model_name_in_literal.go`; registered in `cmd/mxcli/cmd_lint.go`; tests `mdl/linter/rules/mpr018_model_name_in_literal_test.go`", "insight": "When a rule scans strings for model names, the dangerous direction is the one that looks safest. Reflection over ALL string fields is comprehensive and wrong — it cannot distinguish a literal from a reference field, so it reports every microflow call; an allowlist of field NAMES is the safe default because missing an expression field only costs a finding while missing a reference field invents one. The same asymmetry decides literals-versus-expressions. For the false-positive test, one check kills nearly all of it: require the dotted prefix to be a module the project actually has, which excludes prose, file paths, URLs, CSS selectors and Java packages (com.mendix.core.Core) without enumerating any of them. And measure precision on a corpus the rule can actually see: ctx.Microflows() is platform-filtered, so a run over testdata/pedapp mostly measures nothing — the 900-unit Studio Pro-authored submodule testdata/testapp is the real sample. Proving the premise took its own control: mxbuild reported 28 errors before the rename and 28 after, naming the renamed flow in neither, which is what establishes that no checker sees the text reference. `mxcli check