Skip to content

Cockpit decide cards, evidence, and update-instead-of-keep routing for edited Marketplace modules - #10

Merged
MendixMau merged 16 commits into
mainfrom
feat/decide-cards
Nov 16, 2026
Merged

MendixMau merged 16 commits into
mainfrom
feat/decide-cards

Conversation

@MendixMau

Copy link
Copy Markdown
Owner

What

Today's cockpit and pipeline work from the WM1 cold test (Workplace Manager 2.3.1, 9.24 → 10.24 → 11).

  • Cockpit: Decide now cards, one dialog per open question, and approval without a typed name. Each plan question shows the elements, files or findings behind it (evidence.py). The cockpitcheck.py gate fails a page that contradicts itself (ghost steps, a stop bar with nothing to ask, a card without evidence, "for you now" missing a Blocked on you card).
  • Update instead of keep (editroute.py): a Marketplace module the update refused is no longer left on its old version. Each one gets a route:
    • diagram-only edits: update
    • pages, layouts and constants: update, then replay the saved edits
    • unreadable elements: update, then replay
    • theme or JS files: accept the update
    • a dropped vendor attribute nothing else uses: accept the drop
    • other modules' associations: replay them with the original GUID
    • real logic or Java edits: held until moved to <Module>_Ext
  • proposal.py recommends the update. cockpitcheck fails a "keep" recommendation that has no reason and no end date.
  • Dojo to React on every upgrade, and containers on a Rosetta VM on Apple Silicon (earlier commits on this branch).

Tests

  • Full control suite: 2447 passed, 52 failed, 66 skipped.
  • 48 of the 52 failures fail the same way on the previous commit (host tools such as BSD sed, rsync, and docker). The other 4 were mine and are fixed (rerun: 125/127; the 2 left are the release-impact pair, which fail identically on the old commit).
  • New controls editroute-* and cockpit-check-keep-recommended-catch: 14/14 pass.
  • bin/private_names.sh: clean.

Not yet

  • Automated move of real logic edits to <Module>_Ext.
  • Atlas layout edits are replayed in place, not yet moved to the design module.

🤖 Generated with Claude Code

MendixMau and others added 16 commits November 16, 2026 11:08
…local server

Cold test WM1: the plan questions were two or three clicks deep in drawers, never asked as a question,
and the page buttons could not save on the local cockpit.

- Decide now at the top of the live page: one card per open item, one button per option, the
  recommended one marked, the sign-off last; the sidebar and the red bar link to it.
- cockpitserve.py POST /decide: JSON only, through decisions.py import and apply (the same ledger as
  the chat path; a sign-off carries the current plan hash), rewrites the page, appends .cockpit-decided
  for the session that watches it.
- A sign-off stop (exit 14/15) shows a short red bar and no stop rows; the plan choice no longer lists
  the recommended option twice.
- Control cockpit-decide-cards-pass; steps/decide.md documents the page path and the 15 s watch.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…stion with context; approval without a typed name; legend for grey cells

Maurits on the first version: the Decide now panel was too big and still gave too little context, and a typed
name on the sign-off makes no sense when the dev work is tracked in Git anyway.

- Sidebar: Blocked on you, one row per open question with a Resolve button; the dialog says what the item is,
  why it is asked, what each option does and why one is recommended, in words (no file or rule names).
- The sign-off is Approve and continue; cockpitserve.py records the machine's Git name.
- Legend: what Not yet, Not started and Not needed mean.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…rson items never block

Cold test WM1: the plan reached approval while the release-note check had not run
(no brain version tables), the four "needs a person anyway" items showed as open
approvals, the Java cell said "Not yet" after triage scanned, the drawer text ran
off screen, and a cell said only "Found" or "Approved".

- proposal.py: a missing release-impact.md becomes a section-3 question
  (run the check first / go on without it / pause), never a silent pass
- cockpitmatrix.py: person steps show "For a person" / "Later, by a person";
  Java cell reads jarclash-before.md and javafinal.md; a short detail line under
  each cell label; every drawer starts with what we found / what it means /
  what the plan does and why; drawer text wraps; Mendix 11 findings wait for
  the 10 -> 11 step; agent choices that need an OK join Blocked on you
- cockpitpage.py, upgrade-pipeline.sh, decide.md: approval wording, no typed name

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…d grey without the hosted page's database)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ou at once (the poll skips while a dialog is open)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ed plan hid the Approve button)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…nt in the chat

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…tons and journeys, APIs, jobs, widgets, Java); sidebar splits open topics into for you now / the agent closes later / at deploy

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…ntrol reads the for-you-now split

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
….sh); a tool crash in stepupdate stops instead of keeping the module

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…s 3.0.1 package lacks the EXTRA markers Studio Pro writes back)

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
… propose for older runs); the plan lists each widget move as toolkit work, only widgets without a React path are a decision

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…, add cockpitcheck gate

Cold test WM1: after a proposal rerun, 12 "Keep or update module" rows still said
"Not approved yet" with nothing to approve, and the red stop bar stayed after sign-off.

- steps.withdraw: plan-item steps the rerun proposal no longer holds are withdrawn and leave the page
- cockpitpage: no stop bar while the run is running again
- cockpitcheck.py: after every render, flags ghost steps, approval asks with nothing asking,
  a stop bar with no question or while running, and a mismatched "for you now" count
- controls: cockpit-check-ghost-catch, -withdraw-pass, -running-no-stopbar-pass

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Cold test WM1: the update step kept 7 modules at the end of rung 1; kept.txt moved
them from section 2 to section 3, the inputs-hash changed and the whole plan was
asked again. The hash is now taken without kept.txt; the kept modules take their
recommendation (keep old version) under the existing sign-off.
cockpitcheck: a decided plan-item step that the signed PROPOSAL.md does not list is not a ghost.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…idence.py); cockpitcheck fails a card without evidence and compares counts exactly

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
…pdate that carries its edits over (editroute.py)

An edit is no reason to stay on an old Marketplace version (cold test WM1: all seven module cards said "keep").
editroute.py classes each kept module: diagram-only (COSMETIC), UI and constants (REPLAY), unreadable (UNREAD),
files (FILES), dropped vendor attribute nobody uses (VENDOR), other modules' associations (XASSOC, GUID restored).
Real logic and Java edits are held until moved to <Module>_Ext. The pipeline runs it after the update step,
proposal.py recommends the update, cockpitcheck fails a "keep" card without a reason and an end date, evidence
shows the route, and "for you now" lists every Blocked on you card.

Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant