Sync ako/mxcli: widget bindings and action slots, nanoflow expression checks, diag invocation accuracy by ako · Pull Request #1172 · mendixlabs/mxcli

ako · 2026-09-23T14:29:44Z

12 commits from the fork's main, on top of the last sync (c127385). Grouped by theme.

Page and widget authoring

Entity-type JavaScript action parameters (d85c65a, fixes CALL JAVASCRIPT ACTION with entity-type parameter produces wrong BSON type (BasicCodeActionParameterValue instead of EntityTypeCodeActionParameterValue) #1137) — call javascript action wrote BasicCodeActionParameterValue{Argument} where Studio Pro stores EntityTypeCodeActionParameterValue{Entity}, leaving the Studio Pro entity picker empty and failing the build with CE0115. The builder never looked the action up; it hardcoded Basic for every parameter, with the bug stated as a comment. It now resolves the signature and promotes parameters typed to the generic entity <> slot, mirroring the Java-action builder. Measured on 11.6.6: 0 errors fixed, CE0115 faulty, with the two BSON shapes the report quotes on disk. DESCRIBE renders the broken and the correct nanoflow as identical MDL, so the regression test asserts on the $Type rather than on a round trip.
ALTER PAGE can retarget a named action slot (328828c, ALTER PAGE SET cannot address named pluggable-widget action slots (createFileAction etc.) — follow-up to #956 #995) — set 'createFileAction' = microflow M.F on fileUploader1 failed to parse. Named action slots were made writable on CREATE, but alterPageAssignment only accepted an action expression for the key Action, so retargeting one slot meant REPLACEing the whole widget and restating every other property. The mutator refuses a key whose PropertyType is not Action-typed and names the widget's slots, since every WidgetValue carries an Action field — field presence would accept an Integer property and change nothing. Measured on 11.12.1: retargeting DataGrid 2's onSelectionChange keeps PageSize and passes mx check at 0 errors.
Action button captionparams (edfad3f, closes captionparams on an action button passes check and is never written ako/mxcli#632) — accepted by check and not written. Two silent drops: the button builder had its own copy of the template-parameter resolver, which wrote a bare attribute as a literal, and DESCRIBE emitted ContentParams:, which that builder never read — so describe → exec dropped them all and mx check reported CE0720. The shared resolver also brings SourceVariable, association paths and per-parameter format blocks to button captions, and MDL-WIDGET04's orphan-placeholder check now covers them.
A gallery's template block is a slot, not a duplicate name (f275a58, DESCRIBE PAGE output is rejected by mxcli's own check --references (duplicate derived widget names) #978, third symptom) — describe page on a page with several galleries emitted MDL that mxcli's own check --references rejected for a duplicate widget name template1. template and filter are child slots of the gallery's definition: the block's name is discarded on write, which is why DESCRIBE synthesises one per gallery. Measured on 11.12.2: a block authored with a name appears in 0 stored files, and four galleries each holding template template1 pass mx check at 0 errors. Widgets inside a slot still count, and a standalone template — a real container that does store its name — is still checked.
A list view that renders its inputs read-only (0dc5c0f, 1baabaf, List views are written with Editable: false, so every input inside one renders disabled ako/mxcli#631) — a listview that does not say editable: true is written non-editable, and its read-only context wins over editable: Always on the inputs inside it, including inside a nested data view. Valid BSON, clean mx check, clean build; only the running app shows it. The default is kept, since it is Mendix's own — instead MDL-WIDGET31 warns. Editability is read the way the builder reads it, so a quoted editable: 'true' (written false) warns too. Pinned against Studio Pro-authored content at 11.14.0: 30 of 32 list views are stored non-editable and none of those holds an input; the one with inputs was set editable.

`check` catches more before the build

MDL044 over nanoflow bodies and log messages (f9a781f, MDL044 write barrier (currentDeviceType() validation) only runs for CREATE MICROFLOW, not CREATE NANOFLOW — the identical invalid expression passes check, is written, and fails the build with CE0117 #1033) — currentDeviceType() in a nanoflow passed check and exec, then failed the build with CE0117. The earlier fix wired MDL044 to CREATE MICROFLOW only, and log message/node/template expressions were never checked in either flow type — which is where the reported repro puts the call. One list of expression sites now serves both walkers. ValidateNanoflow runs MDL044 only: the full microflow rule set would be a false positive here, since one of its rules refuses something nanoflows allow. Verified on 11.13.0 both ways; the report's own suggested workaround is itself CE0117 in both flow types, and the skill now says so.

DESCRIBE

Java action type-parameter names (7c158bc, DESCRIBE JAVA ACTION drops the type-parameter name (prints entity <> instead of entity <pEntity>) — the declaration does not round-trip #1034) — DESCRIBE JAVA ACTION printed entity <> for a parameter declared entity <pEntity>. The stored type holds only a pointer to the TypeParameter and the reader carried the ID across without resolving the name, which is all the describer prints. The JavaScript action reader already had this pass.

Workflows

jump to inside a boundary-event body (cb52ae7, JUMP TO <activity> inside a boundary-event body writes a Jump with no Target, named after its target — so an interrupting boundary event cannot be expressed correctly #1024) — reported as a jump written named after its target (CE0495 + CE6680). Same root cause as Workflow: dangling jump to target is silently stored as a self-referencing jump (check + exec pass, native mx check fails CE6681) #1005, and already fixed; the report came from a release predating that commit. Measured on a blank 11.14.0 project: 2 errors before the fix, 0 at the fix, 0 on main. The existing tests only put jumps in outcome flows, so this adds boundary-event-body coverage, a bug-test script and a finding. The control is two-sided: on main the test fails only when both guards are stubbed — either alone keeps it green.

`diag loop-report` accuracy

Two defects that made the report describe a different set of runs than the agent actually made.

mxcli runs mxcli (63588cd, closes diag loop-report: mxcli's own child processes are counted as loop calls, and their parent reads as unclosed ako/mxcli#629) — a project reported all 5 test runs as "did not close" while every test passed. test closes its session correctly; the report gave the record to the wrong invocation. Measured: three child processes spawn before the first test executes, and new, eval, tui and the LSP self-spawn the same way. Segmentation closed a parent on a child's start, and the parent's own end — arriving last — landed on whatever was open by then. Two consequences, both fixed: the parent read as a non-zero exit, and the children were counted as calls the agent made, inflating that log's -c and exec counts. Pairing is now by pid with the positional rule kept as a fallback, and the marker rides the environment rather than each spawn site, so it covers all six and any added later. Proven by revert on the measured record shape, with three controls.
A run that fails argument validation (e7862f5, closes A command called with the wrong number of arguments writes no session record at all ako/mxcli#633) — test, check and exec with no arguments wrote no session record at all, because cobra runs argument validation and answers --help/--version before any hook. The session now opens in main(). Two things had to move with it: the close, since --help returns without running the post-hook and would otherwise leave every help lookup reading as a failed run; and the root command's mode, since whoever reaches the singleton first names the session and -c and the REPL were being recorded as neither.

Syntax reference

page.datasource documented a form that cannot parse (45e0ae8, closes syntax page.datasource documents a microflow-argument form that is a parse error ako/mxcli#630) — it printed a microflow datasource with a positional argument; arguments must be named. Found from the parse error rather than from the doc. CLAUDE.md points at mxcli syntax as authoritative instead of restating syntax, precisely so it cannot go stale — which makes a wrong entry the thing consulted instead of checking, and puts the cost in the agent loop. The systemic guard is deliberately not built here, and the reason is measured: 42 of 164 examples fail check, but the large majority are fragments by design, so a blanket "every example must parse" test would be mostly noise until there is a way to mark which examples stand alone.

…ionParameterValue `call javascript action` with an entity-type parameter wrote Microflows$BasicCodeActionParameterValue{Argument: "Mod.Entity"} where Studio Pro stores Microflows$EntityTypeCodeActionParameterValue{Entity: "Mod.Entity"}. The Basic shape leaves the Studio Pro entity picker empty and fails the build with CE0115. addCallJavaScriptActionAction never looked the action up; it hardcoded Basic for every parameter, with the bug stated as a comment. It now resolves the signature through ReadJavaScriptActionByName (already on the backend interface) and promotes parameters typed CodeActions$EntityTypeParameterType — the generic `entity <>` slot — mirroring the Java-action builder, which has drawn this distinction since #656. A parameter typed to a concrete entity keeps the Basic shape, and without a backend the prior behaviour stands. The read, write and DESCRIBE paths already handled the correct type, so the change is confined to the builder. Measured on mxbuild 11.6.6 against testdata/expr-checker/minimal.mpr (which ships NanoflowCommons), two builds of the same script: fixed -> "The app contains: 0 errors." faulty -> CE0115 "The arguments that are passed to JavaScript action 'NanoflowCommons.RefreshEntity' do not match the expected parameters and need to be refreshed." and the BSON on disk carries $Type Microflows$EntityTypeCodeActionParameterValue / "Entity" fixed, versus Microflows$BasicCodeActionParameterValue / "Argument" faulty — the two snippets the report quotes. Control: with the type check stubbed out, the new executor test fails with the reported symptom ("value = *microflows.BasicCodeActionParameterValue, want *EntityTypeCodeActionParameterValue") while both its controls still pass. DESCRIBE renders the broken and the correct nanoflow as identical MDL, so the regression test asserts on the BSON $Type rather than on a round trip. Fixes mendixlabs#1137 Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Ub4kEQmBAN1zk18fjNQppR

`alter page M.P { set 'createFileAction' = microflow M.F on fileUploader1; }` failed to parse (mendixlabs#995): line 2:37 extraneous input 'MyModule' expecting {DROP, ADD, SET, INSERT, REPLACE, '}'} mendixlabs#956 made named action slots writable on CREATE PAGE, but alterPageAssignment only accepted an action expression for the key `Action`, so retargeting one slot meant REPLACEing the whole widget and restating every other property. - grammar: `'<key>' = actionExprV3` and `<key> = actionExprV3`, ahead of the scalar alternatives as on CREATE - executor: an action keyed anything but `Action` routes to the new PageMutator.SetWidgetNamedAction; refused on a column or page-level target, where the setters would stringify it - pagemutator: writes Value.Action (the field widgetobj.Builder.SetAction writes) and refuses a key whose PropertyType is not Action-typed, naming the widget's action slots. Every WidgetValue carries an Action field, so field presence would accept an Integer property and change nothing - MCP backend refuses, as it does for SetWidgetAction Measured on 11.12.1: retargeting DataGrid 2's onSelectionChange via ALTER keeps PageSize and passes `mx check` with 0 errors. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01YVdoySKj31N5PZmjrAGeEh

…5-t3dfxl

) DESCRIBE JAVA ACTION printed `ContextObject: entity <>` for a parameter declared `entity <pEntity> not null`. The stored type holds only a BY_ID pointer to the TypeParameter; javaActionFromGen carried the ID across but never resolved the name, which is all the describer prints. The JavaScript action reader already had this pass. Bare type-parameter references in parameters and the return type are resolved the same way. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011VrrQgJ6aXdEJ8Db1fQx5G

…abs#1033) `currentDeviceType()` in a nanoflow passed `mxcli check` and `mxcli exec`, then failed the build with CE0117 "Error(s) in expression." The mendixlabs#828 fix wired MDL044 to CREATE MICROFLOW only. Separately, `log` message, node and template expressions were never checked for MDL044 in either flow type, and that is where the reported repro puts the call. - checkStmtExprFunctions: the single list of MDL044 expression sites, now including log statements, shared by walkBody and the nanoflow walker - ValidateNanoflow runs MDL044 only. The full microflow rule set would be a false positive here: MDL057 refuses `synchronize`, which nanoflows allow - wired into check (validate_program), the LSP, and exec (validateNanoflowRules in buildNanoflowFromStmt, same allowlist as microflows) Verified on mxbuild 11.13.0: the pre-fix binary writes the repro and the build reports CE0117; the fixed binary refuses it in check and exec. The new sites' positive forms (toUpperCase in a log template, toString([%CurrentDateTime%])) build at 0 errors. The report's workaround `[%CurrentDeviceType%]` is itself CE0117 in both flow types; the skill notes that. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01M1J643ZJw61bNaUSDbZkBj

…s#1024) mendixlabs#1024 reported that `jump to A;` in a boundary-event body was written as a jump NAMED A: [CE0495] "Duplicate name 'A'." at User task 'A', Jump 'A' [CE6680] "The 'Target' property is required." at Jump 'A' This has the same root cause as mendixlabs#1005, and 825873d already fixed it. The report came from v0.20.0, which predates that commit. Measured on a blank 11.14.0 project with native mx check: 825873d^ 2 errors (CE0495 + CE6680, verbatim) 825873d 0 errors main 0 errors The mendixlabs#1005 tests only put jumps in outcome flows, so this adds coverage for the boundary-event body (user task and wait for notification), plus a bug-test script and a finding. Controls: the unchanged test fails with the reported duplicate-name and self-target symptoms against 825873d^. On main it fails the same way only when BOTH mendixlabs#1005 guards are stubbed (jump named after its target, and jumps deduplicated last). Either guard alone keeps it green. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012owAFCYmF2aAUY2S5phmGQ

… duplicate `describe page` on a page with several galleries emitted MDL that mxcli's own `check --references` rejected. mendixlabs#978, third symptom: - duplicate widget name 'template1' (used 4 times) — Mendix requires unique widget names per page (CE0495) The row/col names in the same report were fixed by 587ed31; this one was not covered. `template` and `filter` inside a gallery are child slots of its definition. applyChildSlots stores only the block's children in the slot property and discards the block's name, which is why DESCRIBE synthesises `template1` for every gallery. Measured on 11.12.2: a block authored as `template contentZebra` appears in 0 stored files and describes back as `template1`; four galleries each holding `template template1` pass `mx check` with 0 errors. The check already exempted the enclosing widget's object-list containers; it now exempts its child-slot containers too (definition and every mode), plus the `container <slotName>` spelling that applyChildSlots routes by name. Widgets inside a slot still count, and a standalone `template` (a Forms$DivContainer, which does store its name) is still checked. Control: with the fix reverted, the new test fails with the reported line, `duplicate widget name 'template1' (used 4 times)`. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01Qf5Bk8ZMQkwCNEheapPzq4

…ted them A second test project reported all 5 `test` runs as "did not close" while every test passed, reading it as "the command skips the summary record on success too". It does not: `test` returns normally, PersistentPostRun fires, and the session_end is written. The report gave it to the wrong invocation. **mxcli runs mxcli.** Measured from a real run: session_start pid=1071 test t.test.mdl --skip-build -p … session_start pid=1079 -p …/minimal.mpr -c DESCRIBE SETTINGS session_end pid=None session_start pid=1086 -p …/minimal.mpr -c SHOW MODULES session_end pid=None session_start pid=1093 exec /tmp/mxtest-runner-….mdl -p … Three child processes before the first test executes; `new`, `eval`, `tui` and the LSP self-spawn the same way. buildInvocations segmented on "next session_end OR next session_start, whichever comes first", so a child's start closed the parent and the parent's own end — arriving last — landed on whatever was open by then. session_end carried no pid, so pairing by process was impossible. The rule called itself exact "for sequential invocations, which is what an agent loop produces"; the tool itself is what breaks that. Two consequences, both fixed: the parent read as a non-zero exit, and the three children were counted as calls the agent made — inflating that log's `-c` (111) and `exec` (180) counts. session_end now carries its pid and pairing is by process, with the positional rule kept as the fallback so older logs still read correctly. session_start carries parent_pid, and the marker rides the ENVIRONMENT rather than each spawn site: a child inherits it through exec.Command, so one os.Setenv in Init covers all six self-spawn sites and any added later without touching them. A spawned run is excluded from the table and from wall time — its seconds are already inside its parent's — and reported on its own line. Proven by revert on the measured record shape: the positional rule gives Invocations=4 (want 1), Unclosed=1 for a parent whose tests all passed, and Wall=3 instead of 10. Three controls: an old-format log with no pids still pairs positionally, a top-level run carrying a pid is not treated as spawned, and an orphan session_end from outside the window is dropped rather than applied to whichever invocation is open. Closes #629 Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MgcYSQrLLbUcnAMaCHpyqQ

…t cannot parse `mxcli syntax page datasource` printed `MICROFLOW Module.MF($P)`. Measured: dataview dv (datasource: microflow M.DS_X($State)) -> line 2:15 no viable alternative dataview dv (datasource: microflow M.DS_X(State: $State)) -> Syntax OK Arguments must be named. Reported independently from a test project, found from the parse error rather than from the doc. CLAUDE.md points at `mxcli syntax` as authoritative instead of restating syntax, precisely so it cannot go stale — which makes a wrong entry there the thing consulted INSTEAD of checking, and puts the cost in the agent loop: read it, write it, fail to parse, diagnose, retry. The systemic guard is deliberately not built here. Measured first: 42 of 164 syntax examples fail `mxcli check`, but the large majority are fragments by design — a microflow body, a widget snippet, an OQL fragment — which are legitimately not standalone MDL. A blanket "every example must parse" test would be mostly noise; making it useful needs a way to mark which examples stand alone. Left in #630 with that measurement. Closes #630 Co-Authored-By: Claude Opus 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01MgcYSQrLLbUcnAMaCHpyqQ

fix(executor): entity-type JS action parameters get EntityTypeCodeActionParameterValue

fix(alter-page): SET can address a pluggable widget's named action slot

fix(describe): resolve java action type-parameter names (mendixlabs#1034)

fix(check): a gallery's template/filter block is a slot, not a CE0495 duplicate

…24-osie9e

A listview that does not say `editable: true` is written Editable false, and its read-only context wins over `editable: Always` on the inputs inside it, including inside a nested data view. Valid BSON, clean mx check, clean build; only the running app shows it. The default is kept: false is Mendix's own (mendixmodelsdk 4.115.0, Pages$ListView.editable defaults to false and _initializeDefaultProperties does not set it). Instead, MDL-WIDGET31 warns on a list view that will be written non-editable while it holds an input not marked `editable: Never`. Editability is read with GetBoolProp, as buildListViewV3 does, so a quoted `editable: 'true'` (written false) warns too. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01266huGoXavWeZvPq8BV7LT

Closes #633. `mxcli test`, `check` and `exec` with no arguments wrote no session_start at all, so `diag loop-report` never saw them: probe before after mxcli test 0 1 mxcli check 0 1 mxcli exec 0 1 mxcli check /nonexistent.mdl 1 1 (control) #617 opened the session from PersistentPreRun, but cobra runs ValidateArgs (and answers --help/--version) before any hook. The session now opens in main() before Execute, named from rootCmd.Find(os.Args[1:]); diag stays excluded. Two things had to move with it: - The close. --help and --version return nil without running PersistentPostRun, so with the earlier open every help lookup would have been left unclosed, which the report reads as a failed run. main() now closes after a nil Execute; os.Exit paths still leave no session_end. - The root's mode. Whoever reaches the singleton first names the session, and PreRun had been recording -c and the REPL as "mxcli" instead of "batch"/"repl", the names invocationVerb falls back to. startSession resolves those, plus "help"/"version" for the root flags. The tests re-run the test binary as main(), the only layer that sees the os.Exit path. Controls: stubbing startSession brings back 0/0/0 while the /nonexistent.mdl control still passes. Moving the close back to PostRun gives 0 session_end for check --help, --help and --version. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01FM9DSqkPmhW5KcLtQLgsXn

…scribe `captionparams` on an action button was accepted by `mxcli check` but not written (#632). Two silent drops: - buildButtonV3 had its own copy of the template-parameter resolver that wrote a bare attribute (`[{1} = Title]`) as the literal 'Title'. It now uses the shared buildClientTemplateParams, like dynamictext and grid columns, which also brings SourceVariable, association paths and per-parameter format blocks to button captions. - DESCRIBE printed a button's parameters as `ContentParams:`, which the button builder never read, so describe -> exec dropped them all and mx check reported CE0720. DESCRIBE now emits `CaptionParams:`, and the builder still reads `ContentParams:` for scripts dumped before this fix. MDL-WIDGET04's orphan-placeholder check now covers action/link button captions, so a `{1}` with no parameter is refused at check time instead of failing the build with CE0720. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_016kS7YrCUmW7uFQ1PK6zjkf

) ako/TestApp (Mendix 11.14.0), authored in Studio Pro: 30 of its 32 list views are stored Editable false and none of those holds an input; the one list view with inputs (Pages.EditableLIstView) was set to Editable true. Its describe output checks clean, deleting `Editable: true` from it fires MDL-WIDGET31, and describe -> exec preserves the property. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01266huGoXavWeZvPq8BV7LT

…uo4i # Conflicts: # CHANGELOG.md

test(workflow): pin `jump to` inside a boundary-event body (mendixlabs#1024)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01M1J643ZJw61bNaUSDbZkBj

fix(check): run MDL044 over nanoflow bodies and log messages (mendixlabs#1033)

Two defects a second test project turned up, and three more filed with their measurements

fix: action button captionparams bind a bare attribute and survive describe

fix: a command called with the wrong number of arguments writes no session record

…uo4i

fix(check): warn when a list view renders its inputs read-only (MDL-WIDGET31)

github-actions · 2026-09-23T14:31:26Z

AI Code Review

Let me analyze this pull request against the review checklist.

First, I'll look at the PR description to understand what it's about:

The PR is titled "Sync ako/mxcli: widget bindings and action slots, nanoflow expression checks, diag invocation accuracy"

Looking at the body, it contains several fixes:

Page and widget authoring

Entity-type JavaScript action parameters (fixes CALL JAVASCRIPT ACTION with entity-type parameter produces wrong BSON type (BasicCodeActionParameterValue instead of EntityTypeCodeActionParameterValue) #1137)
ALTER PAGE can retarget a named action slot (fixes ALTER PAGE SET cannot address named pluggable-widget action slots (createFileAction etc.) — follow-up to #956 #995)
Action button captionparams (closes fix: ALTER STYLING writes design properties on pages and snippets (#631) #632)
A gallery's template block is a slot, not a duplicate name (fixes DESCRIBE PAGE output is rejected by mxcli's own check --references (duplicate derived widget names) #978)
A list view that renders its inputs read-only (fixes ALTER STYLING cannot set design properties on widgets in pages and snippets #631)

`check` catches more before the build

MDL044 over nanoflow bodies and log messages (fixes MDL044 write barrier (currentDeviceType() validation) only runs for CREATE MICROFLOW, not CREATE NANOFLOW — the identical invalid expression passes check, is written, and fails the build with CE0117 #1033)

DESCRIBE

Java action type-parameter names (fixes DESCRIBE JAVA ACTION drops the type-parameter name (prints entity <> instead of entity <pEntity>) — the declaration does not round-trip #1034)

Workflows

jump to inside a boundary-event body (fixes JUMP TO <activity> inside a boundary-event body writes a Jump with no Target, named after its target — so an interrupting boundary event cannot be expressed correctly #1024)

`diag loop-report` accuracy

mxcli runs mxcli (closes Add a no-Docker local run/test path (mxcli run --local) and document it in skills #629)
A run that fails argument validation (closes DESCRIBE emits over-qualified grant member names that fail to re-parse (roundtrip break) #633)

Syntax reference

page.datasource documented a form that cannot parse (closes Daily Digest: 2026-06-02 #630)

Now, I need to review this against the checklist.

Let me go through the checklist items:

Overlap & duplication

Check proposals list above for existing proposals covering the same functionality
Check if the diff introduces code that duplicates existing implementations (look for similar function names, command names, types)
Check doctype test from doing code that duplicates existing implementations

Looking at the diff, I see:

New test files for various fixes (e.g., mutator_named_action_test.go, button_caption_params_632_test.go, etc.)
Many new test files in mdl-examples/bug-tests/ for reproducing the bugs
Changes to existing files to fix the issues

I don't see obvious duplication - these appear to be legitimate fixes for reported issues.

MDL syntax design

The PR doesn't seem to add new MDL syntax - it's mostly fixing existing functionality. Let me check:

For the ALTER PAGE named action slot: This is extending existing ALTER PAGE syntax to support named action slots, which was already supported in CREATE PAGE. This follows the pattern.
For the JavaScript action entity-type parameter: This is fixing how existing CALL JAVASCRIPT ACTION statements are processed, not changing the syntax.
For the button captionparams: This is fixing how existing CAPTIONPARAMS syntax is handled.
For the gallery template block: This is fixing how existing syntax is interpreted, not changing the syntax.
For the listview editable inputs: This is adding a warning for existing syntax.
For the nanoflow MDL044: This is extending an existing check to nanoflows.
For the DESCRIBE Java action: This is fixing how existing DESCRIBE JAVA ACTION output is generated.
For the workflow jump to: This is fixing how existing JUMP TO syntax is processed in boundary events.
For the diag loop-report: This is fixing internal diagnostics, not MDL syntax.
For the syntax reference: This is fixing documentation, not MDL syntax.

So no new MDL syntax is being added - just fixing existing functionality.

Full-stack consistency

Since no new MDL syntax is being added, this section doesn't apply. The fixes are to existing features, and they appear to be properly wired through the stack where needed (grammar changes, visitor changes, executor changes, etc. as appropriate for each fix).

Test coverage

New packages should have test files: I don't see entirely new packages, but there are many new test files for the fixes.
New executor commands should have MDL examples in mdl-examples/doctype-tests/: Not applicable as no new commands.
Integration paths should be tested: Yes, there are integration tests (e.g., the diag_loop_report_test.go has integration-style tests).
No time.Sleep for synchronization: I don't see any time.Sleep calls in the diff.

Security & robustness

I don't see obvious security issues. The changes seem to be focused on correctness.

Scope & atomicity

Each commit appears to do one thing:

d85c65a: Entity-type JavaScript action parameters
328828c: ALTER PAGE can retarget a named action slot
edfad3f: Action button captionparams
f275a58: A gallery's template block is a slot
0dc5c0f, 1baabaf: A list view that renders its inputs read-only
f9a781f: MDL044 over nanoflow bodies and log messages
7c158bc: Java action type-parameter names
cb52ae7: jump to inside a boundary-event body
63588cd: mxcli runs mxcli (diag loop-report)
e7862f5: A run that fails argument validation (diag loop-report)
45e0ae8: page.datasource documented a form that cannot parse

These seem appropriately scoped.

Code quality

I don't see obvious code quality issues. The changes seem consistent with the existing codebase.

Bugs & correctness

The PR is fixing bugs, so this is about whether the fixes are correct.

Let me check a few specific fixes against the checklist requirements:

Entity-type JavaScript action parameters (CALL JAVASCRIPT ACTION with entity-type parameter produces wrong BSON type (BasicCodeActionParameterValue instead of EntityTypeCodeActionParameterValue) #1137):
- The fix resolves the signature and promotes parameters typed to the generic entity <> slot
- Added tests: javascript-action-1137-entity-type-parameter.mdl
- Changes: cmd_microflows_builder_calls.go, cmd_microflows_builder_js_action_test.go, microflow_action_test.go
- This appears correct - it's fixing the builder to properly handle entity-type parameters
ALTER PAGE can retarget a named action slot (ALTER PAGE SET cannot address named pluggable-widget action slots (createFileAction etc.) — follow-up to #956 #995):
- Fixes the issue where set 'createFileAction' = microflow M.F on fileUploader1 failed to parse
- Added tests: 995-alter-page-set-named-action-slot.mdl, alter_set_named_action_test.go, visitor_alter_page_named_action_test.go, mutator_named_action_test.go
- Changes: cmd_pages_builder_v3_widgets.go, cmd_alter_page.go, visitor_alter_page_named_action_test.go, mutator.go, alter_set_named_action_test.go
- This adds support for named action slots in ALTER PAGE, which was missing
Action button captionparams (fix: ALTER STYLING writes design properties on pages and snippets (#631) #632):
- Fixes silent drops where captionparams were accepted by check but not written
- Added tests: 632-button-captionparams-bare-attribute.mdl, button_caption_params_632_test.go
- Changes: cmd_pages_builder_v3_widgets.go, cmd_pages_describe_output.go, button_caption_params_632_test.go
- This fixes both the writing and reading of captionparams
Gallery's template block is a slot (DESCRIBE PAGE output is rejected by mxcli's own check --references (duplicate derived widget names) #978):
- Fixes duplicate widget name issue when describing pages with galleries
- Added tests: 978-gallery-slot-names-not-duplicates.mdl, validate_page_slot_containers_test.go
- Changes: cmd_pages_builder_v3_widgets.go, cmd_pages_describe_output.go, validate_page_slot_containers_test.go
- This correctly treats template/filter blocks as slots rather than widgets with stored names
List view inputs not editable (ALTER STYLING cannot set design properties on widgets in pages and snippets #631):
- Adds warning for list views that will render inputs read-only
- Added tests: 631-listview-inputs-not-editable.mdl, validate_listview_editable_inputs_test.go, validate_listview_editable_inputs.go
- Changes: validate_listview_editable_inputs.go, validate_listview_editable_inputs_test.go
- This adds a diagnostic (MDL-WIDGET31) for the problematic pattern
MDL044 over nanoflow bodies (MDL044 write barrier (currentDeviceType() validation) only runs for CREATE MICROFLOW, not CREATE NANOFLOW — the identical invalid expression passes check, is written, and fails the build with CE0117 #1033):
- Extends MDL044 (unknown function check) to nanoflows
- Added tests: 1033-nanoflow-log-expressions-ok.mdl, 1033-nanoflow-unknown-expression-function.fail.mdl, validate_nanoflow_test.go, validate_nanoflow.go
- Changes: validate_nanoflow.go, validate_nanoflow_test.go, validate_microflow.go (modified to not run full set on nanoflows)
- This correctly applies the expression validation to nanoflows
Java action type-parameter names (DESCRIBE JAVA ACTION drops the type-parameter name (prints entity <> instead of entity <pEntity>) — the declaration does not round-trip #1034):
- Fixes DESCRIBE JAVA ACTION to show type parameter names
- Added tests: 1034-describe-java-action-type-parameter-name.mdl
- Changes: java_read.go, java_read_test.go
- This fixes the describer to resolve type parameter names
Jump to in boundary-event body (JUMP TO <activity> inside a boundary-event body writes a Jump with no Target, named after its target — so an interrupting boundary event cannot be expressed correctly #1024):
- Fixes jump to inside boundary-event bodies
- Added tests: 1024-boundary-event-jump-to.mdl, issue1024_boundary_jump_test.go
- Changes: issue1024_boundary_jump_test.go
- This adds test coverage for the boundary-event case
Diag loop-report accuracy (Add a no-Docker local run/test path (mxcli run --local) and document it in skills #629, DESCRIBE emits over-qualified grant member names that fail to re-parse (roundtrip break) #633):
- Fixes issues with session tracking when mxcli spawns child processes
- Added tests: diag_loop_report_test.go (many new tests), session_start.go, session_start_test.go
- Changes: diag_loop_report.go, diag_loop_report_test.go, main.go, session_start.go, session_start_test.go, lsp_diagnostics.go
- This moves session initialization to main() to catch all invocations, including those that fail argument validation
Page.datasource documentation (Daily Digest: 2026-06-02 #630):
- Fixes documentation to show correct syntax (named arguments required)
- Changes: MDL_QUICK_REFERENCE.md
- This corrects the documentation to match the actual syntax requirement

Now, let me check if there are any violations of the checklist:

Looking at the full-stack consistency requirement: For existing features being fixed, they should already be wired through the full pipeline. The fixes appear to be making sure the existing features work correctly through the pipeline.

For example:

The ALTER PAGE named action slot fix adds grammar rules (in MDLParser.g4), visitor handlers (in visitor_alter_page.go), and executor logic (in mutator.go and alter_set_named_action_test.go)
The JavaScript action entity-type parameter fix modifies the builder logic (in cmd_microflows_builder_calls.go) and adds tests
The button captionparams fix modifies both the builder (in cmd_pages_builder_v3_widgets.go) and the describer (in cmd_pages_describe_output.go)
The gallery template block fix modifies both the builder and describer
The listview inputs fix adds a validator (in validate_listview_editable_inputs.go)
The nanoflow MDL044 fix adds a validator for nanoflows (in validate_nanoflow.go)
The Java action type-parameter fix modifies the reader (in java_read.go)
The diag loop-report fixes modify the session initialization and reporting logic

These all appear to be properly wired where needed.

Let me check if there are any specific issues mentioned in the checklist that I should watch for:

"Check if the diff introduces code that duplicates existing implementations" - I don't see obvious duplication
"Check doctype test files above for existing test coverage of the feature area" - The PR adds test files for the bug tests, which is good
"Flag if the PR re-documents already-shipped features as new" - The PR is fixing bugs, not documenting features as new

One thing I should check: the PR mentions "DESCRIBE renders the broken and the correct nanoflow as identical MDL, so the regression test asserts on the $Type rather than on a round trip." for the JavaScript action fix. This is acceptable because DESCRIBE shouldn't be expected to show internal representation differences that don't affect the MDL roundtrip.

Let me look for any potential issues:

In the diag_loop_report.go changes, I see:

Added PID and ParentPID fields to logRecord
Modified buildInvocations to pair by PID
Added Spawned field to invocation
Modified analyzeLoop to exclude spawned runs from counts and add a separate Spawned count

This seems correct for fixing the issue where mxcli's own spawned processes were being counted as agent invocations.

In session_start.go, the session is now started in main() rather than PersistentPreRun to catch invocations that fail argument validation. This is a good fix for #633.

Let me check if there are any MDL syntax issues:

For the ALTER PAGE named action slot, the visitor test shows:

alter page MyModule.UploadPage {
  set 'createFileAction' = microflow MyModule.ACT_CreateFile on fileUploader1;
};

This follows the standard syntax: SET property = value on widget.

For the button captionparams, the test shows:

"CaptionParams": [{1} = Title]

This follows the property format with colon separators (though in the test it's shown as a map, the actual MDL would be CaptionParams: [{1} = Title]).

For the gallery template block, the issue was that DESCRIBE was synthesizing names like template1 for slots that don't store names, causing duplicate name errors. The fix correctly identifies these as slots rather than widgets.

I don't see any violations of the MDL syntax guidelines:

Uses standard CRUD verbs (CREATE/ALTER/DROP/SHOW/DESCRIBE)
Uses qualified names (Module.Element)
Property lists use (Key: value, ...) format
Colon for property definitions, AS for name mappings (I don't see AS being misused in the changes)
Keywords are full English words
Statements read as English
No keyword overloading apparent
Diff-friendly changes
Consistent with existing MDL patterns

One thing to double-check: in the visitor_alter_page_named_action_test.go, I see:

set 'createFileAction' = microflow MyModule.ACT_CreateFile on fileUploader1;

This uses quotes around the property name, which is correct for property keys that might be keywords or contain special characters. The test also shows bare identifiers working.

Now, let me check if there are any specific checklist items I might have missed:

Under "Full-stack consistency (for MDL features)":

Grammar rule added to MDLParser.g4 (and MDLLexer.g4 if new tokens)
AST node type added in mdl/ast/
Visitor handler in mdl/visitor/
Executor handler in mdl/executor/
LSP wiring (cmd/mxcli/lsp.go) if the feature adds formatting, diagnostics, or navigation
DESCRIBE roundtrip — if the feature creates artifacts, DESCRIBE should output re-executable MDL
VS Code extension (vscode-mdl/package.json) updated if new LSP capabilities added

For the ALTER PAGE named action slot fix:

Grammar: I don't see explicit changes to MDLParser.g4 in the diff, but the visitor test implies it parses correctly. Let me check if there are grammar changes...
Looking at the diff, I don't see changes to mdl/grammar/MDLParser.g4 in the selected files, but the PR description mentions it's grouped by theme and the diff is SELECTED. The visitor test shows it parses, so the grammar must have been updated (possibly in the non-selected parts of the diff).
AST: Likely uses existing ActionV3 or similar AST nodes
Visitor: visitor_alter_page_named_action_test.go shows visitor handling
Executor: alter_set_named_action_test.go shows executor handling
LSP: Not mentioned, but this is about parsing/modifying existing functionality, not adding new LSP features
DESCRIBE: Not directly affected by this fix
VS Code extension: Not mentioned as needing update

For the JavaScript action entity-type parameter fix:

Grammar: No change needed (uses existing CALL JAVASCRIPT ACTION)
AST: No change needed
Visitor: No change needed
Executor: cmd_microflows_builder_js_action_test.go shows the fix in the builder
LSP: Not needed
DESCRIBE: Not directly affected
VS Code extension: Not needed

For the button captionparams fix:

Grammar: No change needed (uses existing syntax)
AST: No change needed
Visitor: No change needed
Executor: button_caption_params_632_test.go shows fixes in both builder and describer
LSP: Not needed
DESCRIBE: cmd_pages_describe_output.go is modified (this is the describer)
VS Code extension: Not mentioned as needing update

And so on for the other fixes.

Since the diff is SELECTED and we're only seeing 19 of 57 files, it's likely that the necessary grammar/AST/visitor changes are in the non-selected parts.

Let me check if there are any obvious issues with the changes I can see:

In session_start.go:

// startSession records the invocation before anything can reject it.
// `diag loop-report` reports "mxcli invocations: N", and burning calls on
// malformed invocations is exactly the agent failure it is opened to see
// (ako/mxcli#617). It runs from main() rather than PersistentPreRun because
// cobra validates Args before any hook: an arity failure used to return
// before anything was logged (ako/mxcli#633).
//

This comment correctly explains why the change was made.

In diag_loop_report_test.go, there are comprehensive tests for the PID-based pairing.

In the listview editable inputs test, I see:

{
			name: "quoted 'true' is written false",
			src: `create page M.P (title: 'P', layout: Atlas_Core.Atlas_Default) {
  listview lv (datasource: database M.Thing, editable: 'true') {
    textbox t (label: 'N', attribute: Name)
  }
+}`,
+		

---
*Automated review via OpenRouter (Nemotron Super 120B) — [workflow source](https://github.com/mendixlabs/mxcli/blob/main/.github/workflows/ai-review.yml)*

claude and others added 30 commits September 22, 2026 10:55

Merge remote-tracking branch 'origin/main' into claude/mxcli-issue-99…

8a3c704

…5-t3dfxl

Merge remote-tracking branch 'origin/main' into claude/keen-cray-8ox0aa

952c180

Merge pull request #621 from ako/claude/keen-cray-8ox0aa

0045982

fix(executor): entity-type JS action parameters get EntityTypeCodeActionParameterValue

Merge pull request #622 from ako/claude/mxcli-issue-995-t3dfxl

eee7166

fix(alter-page): SET can address a pluggable widget's named action slot

Merge pull request #623 from ako/claude/zen-cerf-6msn87

a7189f5

fix(describe): resolve java action type-parameter names (mendixlabs#1034)

Merge pull request #624 from ako/claude/mxcli-issue-978-lflti3

7114f89

fix(check): a gallery's template/filter block is a slot, not a CE0495 duplicate

Merge remote-tracking branch 'origin/main' into claude/mxcli-issue-10…

e76ca06

…24-osie9e

Merge branch 'mendixlabs:main' into main

6e9c563

Merge remote-tracking branch 'origin/main' into claude/stoic-brown-m0…

fadff6e

…uo4i # Conflicts: # CHANGELOG.md

Merge pull request #625 from ako/claude/mxcli-issue-1024-osie9e

54fa768

test(workflow): pin `jump to` inside a boundary-event body (mendixlabs#1024)

Merge origin/main into claude/mxcli-issue-1033-p47yne

fbbe33b

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01M1J643ZJw61bNaUSDbZkBj

Merge pull request #626 from ako/claude/mxcli-issue-1033-p47yne

55ec456

fix(check): run MDL044 over nanoflow bodies and log messages (mendixlabs#1033)

Merge pull request #634 from ako/claude/youthful-planck-x6x8xk

f165914

Two defects a second test project turned up, and three more filed with their measurements

Merge pull request #635 from ako/claude/upbeat-feynman-xuwxob

4d8db8e

fix: action button captionparams bind a bare attribute and survive describe

Merge branch 'main' into claude/inspiring-gates-ea783w

b4ef9e3

Merge pull request #636 from ako/claude/inspiring-gates-ea783w

c4626d8

fix: a command called with the wrong number of arguments writes no session record

Merge remote-tracking branch 'origin/main' into claude/stoic-brown-m0…

24c7df2

…uo4i

Merge pull request #637 from ako/claude/stoic-brown-m0uo4i

d5e1468

fix(check): warn when a list view renders its inputs read-only (MDL-WIDGET31)

ako merged commit 57259e8 into mendixlabs:main Sep 23, 2026
5 checks passed

github-actions Bot mentioned this pull request Sep 24, 2026

Daily Digest: 2026-09-24 #1184

Closed

Navigation Menu

Sync ako/mxcli: widget bindings and action slots, nanoflow expression checks, diag invocation accuracy - #1172

Sync ako/mxcli: widget bindings and action slots, nanoflow expression checks, diag invocation accuracy#1172
ako merged 30 commits into
mendixlabs:mainfrom
ako:main

ako commented Sep 23, 2026

Uh oh!

github-actions Bot commented Sep 23, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

Conversation

ako commented Sep 23, 2026

Page and widget authoring

check catches more before the build

DESCRIBE

Workflows

diag loop-report accuracy

Syntax reference

Uh oh!

github-actions Bot commented Sep 23, 2026

AI Code Review

Page and widget authoring

check catches more before the build

DESCRIBE

Workflows

diag loop-report accuracy

Syntax reference

Overlap & duplication

MDL syntax design

Full-stack consistency

Test coverage

Security & robustness

Scope & atomicity

Code quality

Bugs & correctness

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

2 participants

`check` catches more before the build

`diag loop-report` accuracy

`check` catches more before the build

`diag loop-report` accuracy