Remove explicit pnpm version from workflows to fix version conflict - #7

…d read-only permission allowlist (#1555)

AGENTS.md:
- Drop the stale ".github/copilot-instructions.md mirror" instruction — that
  file doesn't exist, and Copilot (coding agent + CLI, since Aug 2025) reads
  AGENTS.md natively. Single source of truth for all agents.
- Multi-agent discipline: prefer one git worktree per agent/task for physical
  isolation; frame the defensive rules as the shared-working-tree fallback.
- Add rule #7: merge only after remote CI is green, never `gh pr merge --auto`
  (#1475 — auto-merge can land a red PR on shared main).

.claude/settings.json (new, committed/shared):
- Read-only permission allowlist derived from recent transcript usage, so
  parallel agents prompt less. Browser/preview observation tools
  (screenshot/snapshot/logs/network/list) + `agent-browser snapshot|skills`.
- Deliberately excludes mutating/arbitrary-exec calls (preview_eval, click,
  fill, start/stop, pnpm, node, npx) and anything Claude Code already
  auto-allows (git/gh read-only, grep, cat, ls, …).

Co-authored-by: Jack Zhuang <277994282+os-zhuang@users.noreply.github.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

#1821)

Small models (e.g. claude-haiku) sometimes answered a "draw a bar chart"
request with a markdown TABLE — running query_data/aggregate_data and
formatting the numbers — instead of calling visualize_data. This was a
tool-selection problem, not a capability gap: the chart preference was buried
as guideline #7 and competed with guideline #4 ("format with markdown
tables").

- data-explorer-skill.ts: add a prominent "Choosing the right tool" section
  ABOVE the guidelines — chart intent (incl. CN terms 图表/柱状图/折线图/饼图/画图)
  → MUST call visualize_data; never substitute a table; reconcile the
  table-formatting guideline; fix duplicate guideline numbering.
- visualize-data.tool.ts: strengthen the tool description to be imperative
  ("the ONLY tool that draws a chart… you MUST call this, not a table; if you
  already fetched the numbers, still call visualize_data to render them").

Prompt-only tuning — no behavior/contract change. Raises the likelihood the
model reaches for visualize_data on a plain chart request without an explicit
"use visualize_data" nudge.

Co-authored-by: Jack Zhuang <277994282+os-zhuang@users.noreply.github.com>
Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

…ta-protocol name; no rename) (#2431)

Per maintainer decision: keep the already-published `@objectstack/metadata-protocol`
package name. Renaming churns downstream for ~0 benefit. The `protocol` suffix is a
deliberate, low-cost naming exception — the real contract lives in
`@objectstack/spec/api`, not the impl package.

- D10: replace the "rename → metadata-runtime" clause with "name retained; content
  converges to the metadata-management impl".
- Open Question #7: resolved (keep the name; README note to clarify impl-vs-contract).
- Status: rev.8; notes D9 step-1 (interface segmentation) shipped in #2429.

Docs only; empty changeset.

Co-authored-by: Jack Zhuang <277994282+os-zhuang@users.noreply.github.com>
Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

…ions (#3563 PR-2) (#3571)

Closes the largest gap in the #3563 route audit: the whole /actions domain —
the documented way to expose custom server-side operations
(engine.registerAction) — had no SDK expression at all, so every console
hand-rolled fetch for it.

- client.actions.invoke(object, action, { recordId, params }) — POST
  /api/v1/actions/:object/:action. recordId travels in the body, which both
  server URL shapes honor. The path is fixed (not discovery-routed): actions
  is not in ApiRoutesSchema, same precedent as projects' /api/v1/cloud.
- client.actions.invokeGlobal(action, opts) — the wildcard
  /actions/global/:action shape.
- The handler's own business failure surfaces as { success:false, error }
  (the dispatcher's inner envelope), deliberately not thrown, so callers can
  toast it. Unit tests cover URL shape, encoding, body defaults, the global
  variant, and the inner-failure path.

Ledger: all three /actions rows flip to sdk; the gap ratchet drops 27 → 24 —
first real exercise of the PR-1 guard.

Documentation stops lying (audit findings):
- client README: the six phantom methods are gone (meta.getObject,
  views.share, views.setDefault, workflow.approve/reject, ai.chat — replaced
  with the real calls where one exists), and the hand-written "all 13
  namespaces / FULLY COMPLIANT" claims now defer to the CI-enforced ledger.
- CLIENT_SPEC_COMPLIANCE.md: retired to a tombstone. Its verdict was measured
  against DEFAULT_DISPATCHER_ROUTES — a table nothing in runtime consumes,
  missing five of the domains it should have been measuring.
- content/docs/api/client-sdk.mdx: documents the new actions surface;
  the automation row now reflects the real method count.

Pre-existing, untouched: the bare `tsc -p` TS2741 (routeMap missing graphql)
reproduces identically on main — its root is the stale graphql entry in
ApiRoutesSchema, tracked as audit follow-up #7 (spec-major territory).

Verified: client 122 passed (5 new), runtime 653 passed, both ledger guard
halves green, eslint clean.


Claude-Session: https://claude.ai/code/session_01LX9ut3MK3KykE11S9bJmv5

Co-authored-by: Claude <noreply@anthropic.com>

…-empty body (objectui#13386) (#13483)

<!-- claude-projects-attribution: {"github_login":"os-elon-musk"} -->
_Requested by **Elon** · [project
thread](https://claude.ai/code/project/chan_01WedDzV26nniwxv7jC88B9D?thread=cmsg_01WedDzV26nniwxv7jC88B9D3jMyxwR7kHLu9j9TjYFimf)_

Fixes #13386
Fixes #12618

Seat: domain:ui#4 · card thread `session_01AqaieDWuDnBRyh3u8epGx6` ·
claim 6391455338
Ruling-ref: 6369306878 (batch #7 item 4, letter C)
Clause-②: no · Metadata-protocol: semantic (the spec side landed as
#25612)

## What changes

Ruling C: in read-only mode an all-empty section that the page keeps
draws its heading plus ONE "Show N empty fields" row, the same toggle a
partly-filled section uses, instead of one em-dash row per field. Edit
mode is unchanged, and a body whose sections are all empty folds once
more (item 2 of #12618).

| Where | Before | After |
|---|---|---|
| `DetailSection`, all-empty section kept (`hideEmpty: false`, or
omitted where no default resolves it), read-only | heading + one `No
value` row per field | heading + one "Show N empty fields" row; the
toggle reveals the rows and "Hide empty fields" folds them back |
| same, edit mode | every field an editable row | unchanged |
| same, print page | every row prints | unchanged: every row prints, no
toggle (#12717) |
| `hideEmpty: true` all-empty section | renders nothing | unchanged |
| partly-filled section | heuristic fold (#7129 Q2-C) | unchanged |
| detail body (`DetailView` sections) where every section is empty and
at least two are kept | one folded section after another | ONE "Show N
empty fields" row for the body (new `EmptySectionsFold`); revealing it
draws each section folded on its own |

No key, shape or default moved, and no new i18n key: both rows reuse
`detail.showEmptyFields` / `detail.hideEmptyFields`. The `hideEmpty`
descriptions in `@object-ui/types` (`DetailViewSection`,
`RecordDetailsSection`, the zod mirror) and the two
`@object-ui/plugin-detail` registration texts now say what a kept
section draws.

## Readings taken (for the seat's review)

1. **"Kept" includes an omitted key where no default resolves it.** The
ruling names `hideEmpty: false`. On `detail-view`, the `detail-section`
node and the `record:details` direct-`fields` fallback, omitted already
meant the same as `false` (keep), and their descriptions said so. The
fold keeps them equal, so the test is "kept" (`!hideAllEmptySection`),
not `=== false`. On `record:details` authored sections nothing changes
for an omitted key: it still resolves `?? true` and hides.
2. **Copy.** The ruling's 「N 个空字段 · 展开」 is described as "the same fold
the mixed sections use", so the row is that fold's existing copy: `Show
4 empty fields` (en), `显示 4 个空字段` (zh). No new string.
3. **Body fold threshold.** The body folds only when every section that
draws anything is all-empty AND at least two are kept: one kept section
is already one row, so folding it again would only drop its heading.
Section groups (`sectionGroups`) turn the body fold off, as they draw
fields of their own beside the sections. A malformed section (no field
array, or a read that throws) turns it off too, so each section still
renders or fails inside its own boundary (#8497).

## Tests

- New:
`packages/plugin-detail/src/__tests__/DetailSection.allEmptyFold-13386.test.tsx`
— the 「交易说明」 fixture: heading + one "Show 4 empty fields" row and no
em-dash; reveal and fold back; untitled section; `hideEmpty: true`
control; edit mode draws four textboxes and no fold; print page prints
four rows; mixed sections (small and large) unchanged; body fold counts
8 over three sections, reveals each section folded, folds back; controls
for a filled field anywhere, a single kept section, edit mode and print
page; `emptySectionsFoldCount` rules including malformed sections.
- Reverse check: with `DetailSection.tsx` at base, 6 of the new file's
cases go red (the edit-mode, print, mixed and count cases stay green, as
they pin unchanged behaviour).
- Updated pins that read an all-empty kept section's rows, now through
the fold: `record-details.hideEmptyRetired-7129` 4/4(b),
`record-details.emptySectionDefault` (`false` keeps; direct-`fields`
fallback), `detailSectionAuthoredHideEmpty-10485`, the AMPLIFICATION
cases of `DetailSection.emptyArray-8474` and
`DetailSection.emptinessAuthority-8376` (now: the fold counts all four
rows, so the empty value is not counted as filled),
`inlineCredentialGate` (`signature`'s empty sample reveals the fold
before reading the pencil),
`record-details.unresolvedIdentityFailClosed-9054` (the legacy
`fieldName` entry survives as "Show 1 empty field").

## Verification (local)

- `pnpm exec vitest run packages/plugin-detail/ packages/types/` — green
after the pin updates (the first full run surfaced the six pins above;
their files re-run green: 132 passed).
- `pnpm --filter @object-ui/plugin-detail type-check`, `pnpm --filter
@object-ui/types type-check` — clean.
- `pnpm --filter @object-ui/plugin-detail lint` — 0 errors.
- `check:new-line-citations` 0 new · `check:doc-json` ·
`check:doc-fences` · `check:control-bytes` · `check:i18n-keys` ·
`check-changeset-presence` — pass.
- `node scripts/check-governed-queue-guard.mjs --test <diff>` — NOT
GOVERNED.
- NOT MEASURED: `check:doc-snippets` / `check:doc-examples` (need the
whole build; CI runs them). NOT MEASURED: screenshots of the HotCRM
opportunity record at 1280 px (#12618 acceptance 2), reason: no HotCRM
console stack in this container; the 「交易说明」 case is pinned in the DOM
test above.

## Acceptance notes

- The `hideEmpty` describe text on the installed spec pin still carries
the pre-#25612 wording; the next pin move picks up the ruled text.
Nothing here reads that prose.

- [x] Adds no product-specific widget, dialog, hook or string (see
AGENTS.md › Product-neutral renderer)

🤖 Generated with [Claude Code](https://claude.com/claude-code)

https://claude.ai/code/session_01AqaieDWuDnBRyh3u8epGx6

---------

Co-authored-by: Claude <noreply@anthropic.com>