Merge remote-tracking branch 'origin/main' into claude/issue-14945-flow-end-node-refused-outcome

Author: claude

.changeset/approvals-bu-member-org-screen.md

@@ -0,0 +1,9 @@
+---
+'@objectstack/plugin-approvals': patch
+---
+
+Fix: a `department` approver on a seeded business unit no longer routes the approval to another organization's members.
+
+`ApprovalService.expandBusinessUnitUsers` screened the `sys_business_unit` rows with the null-inclusive tenant predicate (#3807 — a seeded unit carries no organization and is admitted on purpose) but read `sys_business_unit_member` with no organization predicate at all, under a system context that carries no tenant either. A seeded unit id exists identically in every tenant, so a `department:<id>` approver on tenant A's request resolved the shared unit and then collected every tenant's membership rows hanging off it — approval authority over A's record, routed to B's users. The member read now carries a strict `organization_id` equality against the directory organization the approver resolves in: the same screen `plugin-sharing` applies to these rows, and the same posture this package already takes for `sys_team_member` and `sys_user_position`.
+
+The screen is strict rather than null-inclusive on purpose. `sys_business_unit_member.organization_id` is filled by REST/session writes but left NULL by seed replay and by elevated system-context writes (tracked in #14570), so a NULL on a membership row means unknown tenancy, not "platform-global", and routing fails closed on it. Declared cost: on a deployment whose membership rows (not merely its units) were seeded or system-written, a `department` approver on a request that carries an organization now expands to nobody — the slot falls to the `department:<id>` literal, the existing `expanded to nobody` warning (#3807) names it, and `onEmptyApprovers` governs the request as for any unstaffed target. The repair is to stamp those membership rows. A request that carries no organization is unchanged, and so is every unit-level screen.

.changeset/cron-dialect-row-names-croner.md

@@ -0,0 +1,7 @@
+---
+"@objectstack/spec": patch
+---
+
+The Expression Protocol dialect table no longer names `cron-parser` as the `cron` engine. That package is not a dependency of any ObjectStack package; the row shipped to authors through the generated reference page (`content/docs/references/shared/expression.mdx`) and pointed them at the wrong library for field counts, alias vocabulary and second-field semantics.
+
+The row now says what the code does: no cron syntax is judged at parse time; `croner` evaluates a cron expression only when `CronSchedule.expression` is scheduled (`toBoundaryJobSchedule` → `CronJobAdapter`, where an invalid pattern is refused); every other cron-typed slot is parsed and reaches no engine; and `@objectstack/formula`'s registered `cron` engine has no caller outside that package. Documentation only — no schema, accept set or behaviour changes.

.changeset/diff-usage-error-to-stderr.md

@@ -0,0 +1,17 @@
+---
+"@objectstack/cli": patch
+---
+
+`os diff` with no path arguments no longer prints its usage error on stdout — in either face.
+
+The refusal sat **above** the command's first `if (!flags.json)`, so the face was still undecided when it ran and it fired in **both**. `printError` plus three `console.log` calls — all four writing to stdout — then `process.exit(1)`. Measured on the published entry `bin/run.js` with `NO_COLOR=1` and the streams captured separately, `os diff --json` and bare `os diff` answered byte-identically: exit 1, **141 bytes of prose on stdout, an empty stderr**, and `JSON.parse(stdout)` throwing on the one stream `--json` reserves for the machine.
+
+The diagnostic now goes to stderr, where the rest of this CLI's diagnostics already go. The 141 bytes moved intact — stdout 141 → 0, stderr 0 → 141. Nothing else moves:
+
+- **the exit code is still 1**, so a consumer branching on exit status sees no change at all;
+- **the wording is unchanged**, both usage hints included, so a human reading a terminal sees the same four lines;
+- **nothing is accepted or rejected differently** — no invocation that worked before fails now.
+
+⚠️ **No error payload is invented on this path.** What a `--json` consumer should *receive* on a refusal is an open envelope question, entangled with `os lint --eval --json`'s bare `{ error }` (no `code`, no `httpStatus`), and it is deliberately left open here — this change settles only that the machine's channel no longer carries prose. `--json` on this path emits nothing on stdout; a consumer must still read the exit status, exactly as it must today.
+
+This is the sibling of the `resolveConfigPath` repair, and a genuinely different site: that one is reached through `loadConfig()`, this one is `diff.ts`'s own usage error, raised before any config work happens. The existing pin drives `os diff` with two paths precisely so the run gets *past* this check, so it could not see this path. A new pin (`diff-usage-error-stream.e2e.test.ts`) drives the bare form in both faces, and carries a structural tripwire: across 62 command modules, 27 of which offer `--json`, `diff` was the only one with a stdout write above its guard, and the tripwire goes red if another arrives.

.changeset/dispatcher-scope-strip-environments-prefix.md

@@ -0,0 +1,19 @@
+---
+"@objectstack/runtime": patch
+---
+
+An environment-scoped URL now reaches a dispatcher domain instead of answering 404.
+
+`HttpDispatcher.dispatch()` reads the scoped-URL prefix in three places — the environment-id hint parser, the OAuth-on-MCP gate, and the scope strip that lets `DomainHandlerRegistry` match the remainder. Only the first had been moved to the ADR-0006 `/environments/` spelling; the other two still matched the retired `/projects/` one. The strip therefore never fired on a real scoped URL, and since the registry matches from the head of the path, every environment-scoped request arriving through the `@objectstack/hono` catch-all — the entry cloud hosts mount, and the only one that hands `dispatch()` a still-scoped path — matched no domain at all:
+
+```
+GET /api/v1/environments/<id>/data/task   ->  404 ROUTE_NOT_FOUND   (now: reaches /data)
+GET /api/v1/environments/<id>/health      ->  404 ROUTE_NOT_FOUND   (now: 200)
+GET /api/v1/data/task          (control)  ->  reaches /data, unchanged
+```
+
+The dispatcher-plugin's own scoped mounts were never affected: they pass a pre-stripped subpath (`${prefix}/environments/:environmentId/automation` dispatches the literal `/automation`), which is why the standalone server showed nothing.
+
+The OAuth 2.1 gate moved with it. An access token is honoured only on the MCP surface, and that test runs against the still-scoped path — so `/api/v1/environments/<id>/mcp` would have reached the MCP domain with its token refused had the strip been repaired alone.
+
+**If you still emit the old spelling**: replace `/api/v1/projects/:projectId/...` with `/api/v1/environments/:environmentId/...`, as `content/docs/api/environment-routing.mdx` has instructed since ADR-0006 D2. That prefix is no longer stripped, and it was never a working alias in the first place: nothing parses `/projects/<id>`, so stripping it discarded the only place the request named an environment and served it from the host default instead. ADR-0006 D2 retired `project` on the API surface with no aliases, so the repair is one spelling in all three readings rather than a two-prefix alternation.

.changeset/evaluated-expression-slot-requires-source.md

@@ -0,0 +1,59 @@
+---
+"@objectstack/spec": minor
+---
+
+feat(spec)!: an evaluated expression slot requires a non-blank `source` — `EvaluatedExpressionSchema`, composed by the `assignment` value envelope (#15430)
+
+<!-- adr-0087: not-required (no-migration-prescription) No authorable key is renamed, retired or re-typed: `source` keeps its name and meaning, and every envelope that carried a non-blank `source` parses byte-identically. The two newly refused spellings — an envelope carrying only `ast`, and a `source` that is blank after trimming — never evaluated on any release (the `ast`-only one faulted at run time with the engine's own "persist `source`" prescription, the blank one with a parse error), and a repo-wide census found no in-repo instance of either, so `objectstack migrate meta` has nothing to rewrite and the remedy is authoring a `source`, which the refusal itself prescribes. -->
+
+**BREAKING** in the accept-set sense, landing in the launch window as `minor`
+(the lockstep convention): on the schemas that type an EVALUATED expression
+slot — today the `assignment` node's value envelope,
+`AssignmentExpressionValueSchema` — an envelope with no `source` the engine can
+evaluate is now **refused at authoring**, where it used to parse, register,
+pass `objectstack validate`, and then fault at run time.
+
+Two spellings of one seam, refused by ONE rule with one message at `source`
+(`EVALUATED_EXPRESSION_SOURCE_REQUIRED`):
+
+```yaml
+assignments:
+  digest: { dialect: cel, ast: { kind: const } }   # `ast` only — no engine evaluates it
+  greeting: { dialect: cel, source: '   ' }         # blank after trimming — parses to EOF
+```
+
+> An expression in an evaluated slot needs a non-blank `source`: the expression
+> engine evaluates `source` (the canonical persisted form of phase M9.1) and
+> cannot evaluate `ast` alone, so an envelope carrying only `ast`, or a `source`
+> that is blank after trimming, would validate and register and then fault at
+> run time. Write `{ dialect: 'cel', source: '…' }`.
+
+- **`ExpressionSchema` is NOT narrowed.** It is the persistence contract —
+  `source` OR `ast` — and its docblock declares that `ast` becomes required in
+  build output at phase M9.2. The new export `EvaluatedExpressionSchema` (and
+  its type `EvaluatedExpression`) is a sibling: the same envelope with `source`
+  required and non-blank, spelled once and composed by every evaluated slot, so
+  when AST-only evaluation lands the flip is one edit there rather than a
+  per-slot unwinding. The rule is worded as "an evaluated slot requires whatever
+  the engine can actually evaluate"; what that is today is `source`.
+- **The notion of blank is the engine's own** — `.trim()`, which
+  `cel-engine.ts`'s helpers already apply — not a third one beside the shape
+  rule's `min(1)` and `validateExpression`'s trim.
+- **Three doors agree.** `registerFlow` refuses the flow, `objectstack validate`
+  and the runtime publish gate report a located `error` at the author's own
+  variable (`config.assignments.<name>.source`), and the executor's own shape
+  pass refuses the same set — all through the spec schema, so none of them
+  grew a rule of its own.
+
+**What an author does with a refused envelope.** An assignment value that
+carried only `ast` has no evaluable form under M9.1: author its `source`. A
+whitespace-only `source` was never an expression: delete the entry, or write
+the expression. Every envelope with a non-blank `source` is unchanged, and
+nothing is renamed, retired or rewritten — the refusal itself carries the
+prescription.
+
+Not touched here: the `predicate` half of the same seam — `evaluateCondition`'s
+silent `false` on an envelope without a `source` — is a behaviour change on a
+live path with its own card, and the edge-condition schema that carries that
+envelope is narrowed in a follow-up once the in-flight change to
+`automation/flow.zod.ts` lands.

.changeset/layer0-verdict-on-operation.md

@@ -0,0 +1,17 @@
+---
+"@objectstack/spec": minor
+"@objectstack/objectql": minor
+"@objectstack/plugin-security": minor
+---
+
+feat(security): the Layer 0 tenant wall records its verdict on the operation, and the bulk data-event producer reads it instead of re-deriving the wall
+
+`BulkDataEventSchema.organizationId` is stamped on a `data.records.updated` / `data.records.deleted` event only when the Layer 0 tenant wall named exactly one organization for the whole predicate write. The producer (`publishBulkDataEvent`, `@objectstack/objectql`) used to decide that by re-deriving the wall's inputs — posture, context, and the object's own tenancy clauses. It could never see the third clause plugin-security folds into `tenancyDisabled`: the deployment-declared `platformGlobalObjects` carve-out (#12699). On such an object under an armed wall the producer stamped the caller's organization while Layer 0 had composed no wall at all — a wrong key asserting "every affected record belongs to this organization" over a batch that could span several, the #13566 leak shape reappearing on the bulk path (#15706).
+
+Ruled on #15706 (seam (i), ADR-0131 D8 「一道谓词，算一次」): the wall records what it decided, and the reader composes nothing.
+
+- **`@objectstack/spec`** — new export `TenantLayer0VerdictSchema` / `TenantLayer0Verdict` (`@objectstack/spec/security`): the four verdicts a Layer 0 wall can reach for one operation — `none`, `organization`, `organizations`, `deny`. Additive.
+- **`@objectstack/objectql`** — `OperationContext` gains an optional member `tenantLayer0Verdict`, written by the enforcement layer at the moment it composes the wall onto the operation's predicate. Additive widening of a published surface, hence `minor`. `publishBulkDataEvent` now reads that member and nothing else: a recorded `organization` (or a one-member `organizations`) verdict stamps the key; `none`, `deny`, a multi-member set, a malformed value, or NO recorded verdict all omit it. The engine no longer consults the enforced posture, the execution context or the object schema to answer the question — the mirror is deleted, not moved.
+- **`@objectstack/plugin-security`** — the engine middleware records `opCtx.tenantLayer0Verdict` on every operation whose predicate it composes the wall onto (reads and predicate writes); `computeTenantLayer0Filter` is now a projection of the new `computeTenantLayer0Verdict`, so the recorded verdict and the injected predicate come from one computation. An on-behalf-of write records the intersection of the caller's and the delegator's walls. System contexts and by-id writes record nothing (no wall is composed for them).
+
+What moves, and in which direction: a deployment-exempted object under an armed wall now publishes `organizationId` ABSENT (it was wrongly present); a `PLATFORM_ADMIN` rung on a PUBLIC tenant object now publishes it PRESENT (the wall stands there; it was conservatively absent); a hand-built context with no rung is answered by the plugin's capability probe rather than conservatively absent. Every population the previous producer answered correctly is unchanged.

.changeset/membership-ended-session-revoke.md

@@ -0,0 +1,41 @@
+---
+'@objectstack/platform-objects': minor
+'@objectstack/plugin-auth': minor
+---
+
+`sys_session.revoke_reason` accepts `organization_membership_ended` — "Remove member" now actually signs the person out
+
+Removing a member deleted the `sys_member` row and left the session alive, for up to seven
+days. #15409 closed the security half per request (a session whose `activeOrganizationId`
+is not backed by a membership resolves with no active organization). This is the courtesy
+half an admin was promised, and it is **never the enforcement**: a trigger can be missed,
+an evaluation cannot.
+
+- **New `revoke_reason` value, `organization_membership_ended`** — an accept-set widening
+  on a published system object, hence `minor` on `@objectstack/platform-objects`. Every
+  reason before it is a timer (`idle_timeout`, `absolute_max`, `concurrent_cap`) or an
+  interactive revoke (`user_revoked`, `admin`); this is the first authorization-event
+  cause. There is no Zod enum behind the column — it is free `text` — so the field's own
+  description is the published vocabulary, and that is where the value is declared. The
+  string deliberately matches the one the API-key arm of the same ruling family already
+  mints for this event (`authRefusal.reason` in `resolve-authz-context.ts`), so one grep
+  finds every place the platform acts on a membership ending.
+- **The trigger acts on the ORGANIZATION'S CLAIM, never on the user** (maintainer ruling,
+  decision batch #49 item 4, option B). A user who still holds another membership is
+  **re-pointed** to it — never signed out of organizations they legitimately belong to. A
+  user with no remaining membership has their session revoked through the existing
+  `revoked_at` / `revoke_reason` mechanism, which expires it in place: better-auth returns
+  nothing on the next request and the Console's existing 401 → login redirect handles it,
+  with **no client change**.
+- **The seam is an engine hook on `sys_member`**, not a hook on better-auth's
+  `/organization/remove-member`. A census measured that the endpoint, a direct delete, a
+  bulk delete, the cascade from a `sys_user` delete and an organization re-point all reach
+  the hook, while an endpoint hook would have reached one of them. Same precedent as
+  `last-admin-guard.ts`.
+- **New public surface on `@objectstack/plugin-auth`** — `MEMBERSHIP_ENDED_REVOKE_REASON`,
+  `endSessionClaimsForEndedMembership` and `registerMembershipEndedSessionTrigger`, hence
+  `minor` rather than `patch`.
+
+Known open by measurement, not by omission: a raw driver delete bypasses the trigger
+entirely, and cloud's package-uninstall sample-data purge is one (filed as cloud#2003). The
+per-request check covers it; the courtesy does not.

.changeset/olive-spiders-refuse.md

@@ -0,0 +1,13 @@
+---
+"@objectstack/cli": minor
+---
+
+**BREAKING** `os create <type> <name>` now refuses a project name that npm refuses, and refuses it before it writes anything.
+
+`os create plugin "My App"` used to exit 0 having written `./plugin-My App/`, carrying a manifest that read `name: "@objectstack/plugin-My App"`. Nothing failed at scaffold time, so the invalid name surfaced later at `npm publish`, in the terminal of whoever ran it next. `os init` has always refused that same input before touching the disk. The rule set is now shared between the two scaffolders rather than restated in one of them, so they answer the same way.
+
+`os create` also refuses a name whose composed scoped package name exceeds npm's 214-character ceiling. `@objectstack/plugin-` spends 20 of those characters before the name begins, so a name that `os init` accepts can still compose to one npm rejects; that check sits next to the composition rather than in the shared rule set.
+
+A scripted invocation that passed an invalid name now exits 1 with the reason on stderr, where it previously exited 0 and produced a project that could not be published.
+
+<!-- adr-0087: not-required (no-migration-prescription) The change narrows what a CLI argument accepts at invocation time. No metadata surface, stored row or spec declaration is touched, so `objectstack migrate meta` has nothing to carry and the ledger has nothing to record. -->