fix(plugin-auth, runtime): four principal-less producers take the explicit system opt-in

[objectstack-fleet]

Fixes #21912
Clause-②: no

This PR is a slice of #21908, the closure of the security middleware's principal-less hand-off (ADR-0096). It covers the identity and runtime producers. #21908 stays open for the deny, which lands last.

What moved

Each producer below reached the data engine with no principal and no isSystem. That is the hand-off, and it is not an authorization. Each one now takes the explicit system opt-in that already exists. ⛔ No new elevation API, no change to what any door authorizes, no accept-set change.

Row	Position (function)	Engine calls	Route taken
17	plugin-auth auth-plugin.ts, the platform-admin OAuth client toggle route (/admin/oauth2/toggle-disabled)	findOne + update sys_oauth_application	withSystemContext, the wrapper better-auth's adapter already writes these rows through
18	plugin-auth scim-connection-service.ts verifyScimBearerToken	findOne sys_scim_connection_credential	isSystem: true in the read's trailing options
19	plugin-auth auth-manager.ts organizationHooks.beforeUpdateOrganization (the slug guard)	findOne sys_organization, find sys_environment	withSystemContext
21	runtime http-dispatcher.ts enforceProjectMembership	find sys_environment_member	isSystem: true as the read's query context

Row 20 moves nothing. I read every site, and each one already runs with the opt-in:

• adopt-membership.ts adoptExistingMembership: its only caller hands it the adapter's withSystemContext engine.
• membership-ended-session.ts endSessionClaimsForEndedMembership: all four calls pass { context: SYSTEM_CTX } (isSystem: true) as the trailing options, and the engine honours that argument on reads and writes.
• The auth-manager.ts insert helper (settleSelfRegistrationGrant, with findPermissionSetRows): it reads and writes through withSystemReadContext, the deprecated alias of withSystemContext.

The auth-manager.ts edit (row 19) was made after #21872 landed, on a merge of origin/main that contains it.

Measured: no gate fires on any moved call today

An isSystem context short-circuits the gates the hand-off still runs before next(): package-managed, system-row, curated-capability, audience-anchor, engine-owned and delegated-administration. A move is neutral only if none of them fires on the producer's calls.

• Static. Each gate is keyed to objects and verbs that none of these calls touch. The first four guard writes to sys_permission_set, sys_position, sys_capability and sys_position_permission_set. Engine-owned needs a userId. Delegated-administration guards writes to the RBAC link tables, sys_permission_set and sys_member. Rows 18, 19 and 21 are reads. Row 17 writes sys_oauth_application, which none of the gates names.
• Instrumented. I added a local, uncommitted probe in security-plugin.ts. It recorded each principal-less, non-system context that reached the hand-off, with its stack, and each gate refusal of such a context. Over every run below it recorded 0 gate refusals. Every call of the card's functions reached the hand-off, so no gate had stopped it.

Per function, before → after (records at the hand-off):

Function	dogfood subset	dev boot	runtime harness
toggle route (row 17)	5 → 0 (findOne 3, update 2)	5 → 0	—
verifyScimBearerToken (row 18)	0 → 0	1 → 0	—
beforeUpdateOrganization (row 19)	0 → 0	1 → 0 (sys_organization findOne)	—
enforceProjectMembership (row 21)	—	—	2 → 0
row 20 functions	0 → 0	0 → 0	0 → 0
all records	1601 → 1596	300 → 293	39 → 37

Before = the base tree with the probe. After = the change with the probe: the dev boot and the harness on the final tree (9878b925), and the dogfood subset on the pre-merge commit 4239dd47, whose row 17 code is the same. The boot's background ticks (the outbox claims) make the totals differ by a few records between runs. The per-function counts are the reading.

• Dogfood subset. Seven files that reach the card's functions: the two platform-admin route sweeps, the organization-update door, the two SCIM-enabled suites, org-admin reach and membership attribution. 57 tests passed both times. Only row 17 appears in the dogfood suite. This subset reproduces the full-suite census of the measure-first round (6003676228) for these rows exactly.
• Dev boot. pnpm dev -- --fresh on showcase with SCIM enabled, driven as the seeded admin. It registers an OAuth client, toggles it twice, toggles a missing id, sends a SCIM request with an unknown bearer, and changes the default organization's slug. The answers were identical before and after: register 201, toggles 200 / 200 / 404, SCIM 401, slug update 200.
• Runtime harness. A scratch file, deleted afterwards, booted a real engine with SecurityPlugin and called enforceProjectMembership for a member and a non-member. No open-source composition reaches row 21: no KernelResolver sets environmentId, and sys_environment_member is a cloud control-plane object. The answers were null and 403 both times.
• Restored. The probe was reverted (security-plugin.ts blob 5b4ab280 equals HEAD), plugin-security was rebuilt, and ablation-dist-preflight --absent confirms the marker is gone from dist/. The positive control: 4 hits in dist/ while the probe was live.

One difference that is not a gate (row 17). Under the hand-off, the engine's static read-only strip ran on the toggle's update and dropped the updated_at the route supplies, with a WARN. Under isSystem the strip does not run. I compared the stored rows: on the SQL driver, both paths store disabled and an updated_at equal to the driver's own stamp. The only change is that the WARN line no longer appears on each toggle.

Pins (one per package) and ablations

• plugin-auth/src/principal-less-producers-system-context.test.ts: a real engine with a context-recording middleware. The toggle route's findOne and update and the SCIM probe's findOne are isSystem. The route still answers 200 and flips the stored flag, and still answers 404 RESOURCE_NOT_FOUND. The verifier still resolves a known bearer to its connection, and still answers null for an unknown one.
• plugin-auth/src/auth-manager.org-slug-guard-system-context.test.ts: both slug-guard reads are isSystem, and the guard still refuses with FORBIDDEN / 403 while an active environment exists. On an engine that refuses a principal-less, non-system context, the guard still refuses. The pre-existing catches are pinned as they stand: a read that throws ends the hook without refusing.
• runtime/src/http-dispatcher.membership-system-context.test.ts: the membership read is isSystem, a member passes, and a non-member gets 403 PROJECT_MEMBERSHIP_REQUIRED. On an engine that refuses a principal-less, non-system context, the non-member is still refused. The pre-existing fail-open catch is pinned as it stands: a read that throws lets the request through.
• Ablations. Each went through scripts/ablation-replace.mjs: the anchor hit once, the mutation was verified on disk, and the restore was proven (blob equals HEAD, git diff HEAD empty). Each pin imports its subject from src, so no build sat between the mutation and the run.
  • A, row 17, withSystemContext dropped: 2 red.
  • B, row 18, trailing context dropped: 2 red.
  • C, row 21, query context dropped (re-run on 9878b925): 3 red, including the refusing-engine non-member case.
  • D, row 19, withSystemContext dropped: 2 red, including the refusing-engine case.

Tests and gates (at 9878b925)

• New pins, on 9878b925: plugin-auth 2 files, 9/9 passed. runtime 1 file, 5/5 passed.
• pnpm --filter @objectstack/runtime exec vitest run --project local --maxWorkers=2 on 9878b925: 330 files, 4654 passed, 19 skipped. pnpm --filter @objectstack/runtime run typecheck: exit 0.
• plugin-auth on 60c5f22c: the suite (126 files, 2607 passed, 10 skipped) and run typecheck (exit 0). The only commit since, 9878b925, touches runtime and the changeset, and plugin-auth imports neither.
• The runtime suite caught a spelling of mine. The membership read first carried its context as a trailing third argument. Eight existing assertions read the read's two arguments: toHaveBeenCalledWith in http-dispatcher.test.ts and http-dispatcher.membership-skip-boundary.test.ts. They turned red. The context now rides inside the query instead. The opt-in is the same and so is the engine's reading (ObjectQL merges the two), and both suites pass unedited.
• eslint --no-inline-config over the 7 changed .ts files: 7 files, 0 errors, 0 warnings. These 7 are the whole population whose lint verdict this diff can move. eslint.config.mjs never enables type-aware linting (no parserOptions.project, no typed rules), so no untouched file's verdict can change. The repo-wide pnpm lint is CI's.
• node scripts/pm/dispatch-gates.mjs --commands --repo objectstack-ai/objectstack on 9878b925 derived 98 commands. All 98 ran and exited 0. The --ran reconciliation over the exit-coded record reads: "98 derived famil(ies) accounted for — 98 run, 0 NOT-MEASURED (a DERIVED zero — all 98 recorded an exit code and none of them is 3)". The same 98 also ran green on 60c5f22c.
• The branch sits 3 commits behind origin/main (9dce6353). Those commits touch content/docs/permissions/sso.mdx and a rest test, none of this diff's files. fix(plugin-security): the packaged-permission-set lock refusal carries its guidance as userMessage #21902 is merged into main and contained in this branch.

Acceptance notes

• The fail-open catches on rows 19 and 21 are unchanged. They are pre-existing, and row 21's is documented as deferred. This PR removes the path by which a principal-less deny would trip them: both reads are now isSystem. A read that throws for any other reason still skips the slug guard (row 19) or opens the membership gate (row 21). Both behaviours are pinned as they stand, so the seat can sequence them before the deny.
• Row 19 in the open-source composition. sys_environment is not registered there, so the environment read throws before it reaches the engine middleware. The catch then ends the hook, and the slug guard never refuses in an open-source deployment. It acts only where the object exists. Measured on the dev boot: the slug change answered 200 and recorded no environment read at the hand-off.
• NOT MEASURED: a cloud composition. An isSystem read also bypasses any host read hook keyed on the caller, such as a control-plane org-scope hook. I measured the six named gates only, and only in-repo.
• mintScimConnectionCredential inserts without the opt-in. It has no runtime caller (tests only) and is not exported from the package entry, so nothing produces through it today. Noted, not changed.
• The census page (content/docs/permissions/system-context.mdx) is current. --fix moved its held declaration count from 25 to 26, for the new trailing-options type on the SCIM probe. It asked for no anchors.

---

Generated by Claude Code

[github-actions]

📓 Docs Drift Check

This PR changes 2 package(s): @objectstack/plugin-auth, @objectstack/runtime, touching 7 documentable anchor(s).

15 hand-written doc(s) NAME something this change touched and may need an implementation-accuracy re-verification:

• content/docs/api/data-flow.mdx (via findOne (symbol, a method of interface CredentialEngine))
• content/docs/api/environment-routing.mdx (via enforceProjectMembership (symbol, a method of class HttpDispatcher))
• content/docs/automation/hook-bodies.mdx (via findOne (symbol, a method of interface CredentialEngine))
• content/docs/automation/webhooks.mdx (via findOne (symbol, a method of interface CredentialEngine))
• content/docs/kernel/contracts/data-engine.mdx (via findOne (symbol, a method of interface CredentialEngine))
• content/docs/kernel/contracts/index.mdx (via findOne (symbol, a method of interface CredentialEngine))
• content/docs/kernel/events.mdx (via findOne (symbol, a method of interface CredentialEngine))
• content/docs/permissions/attachments-access.mdx (via findOne (symbol, a method of interface CredentialEngine))
• content/docs/permissions/field-level-security.mdx (via findOne (symbol, a method of interface CredentialEngine))
• content/docs/permissions/record-view-auditing.mdx (via findOne (symbol, a method of interface CredentialEngine))
• content/docs/permissions/rls.mdx (via findOne (symbol, a method of interface CredentialEngine))
• content/docs/permissions/system-context.mdx (via findOne (symbol, a method of interface CredentialEngine))
• content/docs/protocol/objectql/query-syntax.mdx (via findOne (symbol, a method of interface CredentialEngine))
• content/docs/protocol/objectql/schema.mdx (via findOne (symbol, a method of interface CredentialEngine))
• content/docs/ui/react-pages.mdx (via findOne (symbol, a method of interface CredentialEngine))

⛔ 6 release-owned page(s) also name something this change touched. These are read-only:

• content/docs/releases/v15.mdx (via findOne (symbol, a method of interface CredentialEngine))
• content/docs/releases/v16.mdx (via findOne (symbol, a method of interface CredentialEngine))
• content/docs/releases/v17/17-0.mdx (via findOne (symbol, a method of interface CredentialEngine))
• content/docs/releases/v17/17-5.mdx (via findOne (symbol, a method of interface CredentialEngine))
• content/docs/releases/v17/17-6.mdx (via findOne (symbol, a method of interface CredentialEngine))
• content/docs/releases/v17/index.mdx (via findOne (symbol, a method of interface CredentialEngine))

content/docs/releases/ is RELEASE-OWNED (AGENTS.md "Documentation Guardrails"): release
notes are written centrally at release time, and a code PR that edits them is the exact PR
that guardrail exists to stop. They are still audited — read-only. If one of them is actually
wrong, file an issue or open a dedicated docs-only PR; do not edit it here.

What this run could not see

• the SDK route bridge reached 54 of 206 client-bound route-ledger rows — the other 152 have no registrar path: tail to select them, so pages documenting THEIR client methods cannot appear above, on this or any run. Of those 152: 0 are remediable by widening that discovery convention (an in-repo file declares the path; the convention did not scan it); 55 are structural — on a ledger where NOT ONE row is declared in-repo, so no discovery change reaches them at any price; 97 are undecided (no in-repo declaration, on a ledger that has other in-repo registrars — absence and an unreadable spelling are not distinguishable here). The rows themselves: node scripts/docs-audit/affected-docs.mjs --bridge-coverage
• a page that states a rule by its inputs shares no identifier with the emitter that implements the rule, so an emitter-only diff cannot list it — not on this run and not on any run. Measured on fix(driver-sql): emit varchar(maxLength) for a text field a declared index keys on #11430: content/docs/protocol/objectql/types.mdx documents the text-family column mapping by the ObjectQL type names it maps FROM (text / textarea / html) while the diff changed createColumn; it went unlisted, and it was the page that diff falsified, in four places. No shared token exists to detect this on, so a rule your change carries has to be re-read by hand in the pages that restate it.
• a key NAME is not a key, so the hand re-read the line above prescribes can land on the wrong schema. The same spelling is authorable on one governed type and a [REMOVED] tombstone on another for each of active, aria, joins, objects, template, tools and version (censused on [finding] tools is a key on BOTH AgentSchema (tombstoned, dead) and SkillSchema (live, cloud-attested), so a name-based search attributes skill examples to the agent key — it produced a false stop-the-line alarm on PR #19059 #19093 over the liveness ledger's governed types, top-level keys); nothing in a search result distinguishes the two, so a grep hit on a LIVE example reads as evidence about the DEAD key. Measured on fix(spec): the agent.tools liveness row says dead — it claimed live on a key the schema tombstoned #19059: content/docs/ai/agents.mdx was reported as contradicting the agent.tools tombstone over its tools: example at :161, which is inside the defineSkill({ block opened at :155 — the page was already correct. Settle ownership by PARSING the value against both schemas, never by the name: that literal PASSES SkillSchema, and as an AgentSchema it FAILS at tools with the tombstone prescription. ⛔ These names are not the whole class — a key retired through a .strict() guidance map leaves no tombstone in the walked shape and none of them here (tool.category, live as AIToolDefinition.category).

Coarse fallback — 36 page(s) merely mention a changed package (the pre-#9192 predicate, kept for the deliberately-wide backstop): node scripts/docs-audit/affected-docs.mjs --json 9dce635337c2cc42a4149aa49289ad77d172363d → packageMentionDocs.

Which tree this was computed on

This run read content/docs from 9378e862d9522e8d4b3b335b064f391cbfa7dde0 — the merge of head 9878b925fc207120aae9966a26929a58d274c268 into base 9dce635337c2cc42a4149aa49289ad77d172363d, which is what actions/checkout gives a pull_request run. Not the PR head.

A worktree cut from an older main holds a different content/docs, so re-deriving there can legitimately return a different list — that is a different tree, not a wrong row. To answer on the same tree:

# while this PR is open — GitHub drops the merge commit once it closes
git fetch origin 9378e862d9522e8d4b3b335b064f391cbfa7dde0 && git checkout 9378e862d9522e8d4b3b335b064f391cbfa7dde0
# afterwards, rebuild it from the two parents, which stay fetchable
git fetch origin 9dce635337c2cc42a4149aa49289ad77d172363d 9878b925fc207120aae9966a26929a58d274c268 && git checkout -B drift-repro 9dce635337c2cc42a4149aa49289ad77d172363d && git merge --no-ff 9878b925fc207120aae9966a26929a58d274c268

node scripts/docs-audit/affected-docs.mjs --json 9dce635337c2cc42a4149aa49289ad77d172363d

⚠️ That checkout carried uncommitted changes, so the commit above does not fully identify what was read.

Advisory only, and a precision-first one (#9192): a page is listed because it names a
symbol, wire route or SDK method this diff touched — not because it mentions a changed
package. Each row says which anchor put it there, so a wrong row is reportable rather than
merely annoying. To re-verify, run the docs-accuracy-audit workflow scoped to these files:
node scripts/docs-audit/affected-docs.mjs 9dce635337c2cc42a4149aa49289ad77d172363d → pass the list as
args.docs, on the commit named under Which tree this was computed on.