@sema-agent/sdk 16.2.0 → 17.0.0

package/openapi.yaml

@@ -417,7 +417,7 @@ paths:
         prepare-time refusal code that HAS a status in the server's per-code triage table (`resume_at.*` — `resume_at.not_found` is 404, the rest 422 —, governance codes,
         `rewind.store_unconfigured` / `rewind.capture_disabled` / `rewind.child_scope_unsupported` / `rewind_snapshot.unresolvable`, the
         registered `config.*` codes incl. the three roster-bound ones and — server >= 7.116.0 — `config.invalid_checkpoint_store` (400) and
-        `config.user_invoked_skill_unknown` (400: top-level `userInvokedSkills` named a skill this run does not carry), `corrupt_session`; codes outside the table, e.g. `rewind.invalid_spec`, stay
+        `config.user_invoked_skill_unknown` (400: top-level `userInvokedSkills` named a skill this run does not carry) and — server >= 7.117.0 — `config.image_invalid` (400: an inline image failed the engine's byte admission; `error` starts with `TaskSpec.images: `), `corrupt_session`; codes outside the table, e.g. `rewind.invalid_spec`, stay
         200 with a failed body) answers its 4xx / 5xx ERROR BODY only when the whole submission did ZERO WORK (cumulative
         `stats.turns === 0` AND `stats.tokens === 0` — the common case: nothing billed, not idempotency-cached). When a
         `verify` / `cascade` run's earlier tier already ran turns and only the last tier is refused, the SAME code arrives
@@ -869,7 +869,9 @@ paths:
             evicts an already-accepted instruction to make room. Opposite handling from not_running: the
             text is still valid — let the run resume (which drains the queue) or reduce concurrent steers,
             then resend.
-            errorCode "steering.duplicate_input_id" (server >=7.3.0 / core 5.14.0) — the Idempotency-Key you
+            errorCode "steering.duplicate_input_id" (server >=7.3.0 / core 5.14.0; server >= 7.118.0 adds `detail.differs`, a
+            non-empty subset of SteerReplayFacet — contains `text` ⇒ fresh key; otherwise the same instruction, replay with the
+            original values and do NOT change the key) — the Idempotency-Key you
             sent is already parked with DIFFERENT steering content (same key + same content is an idempotent
             no-op and 202s). The engine refuses rather than silently picking a side. Reissue with a fresh
             key — this is a new instruction, not a retry.
@@ -886,6 +888,13 @@ paths:
             ⚠️ Against a server 7.19.0–7.22.0 the OLD caveat applies verbatim: only the parked legs dedupe —
             a lost 200 retried with the same key injects TWICE with byte-identical receipts; treat a lost 200
             there as "outcome unknown", never "safe to retry".
+            🔴 server >= 7.117.0 — TWO MORE 409 answers on a TERMINAL run (the `parked_for_wake` leg), refused BEFORE the wake card is minted
+            (no card is minted; the row is untouched; on 7.116.x this was a 202 `parked_for_wake` followed by a refusal at `/wake`, i.e. a card that could
+            never be redeemed): when the stored submit body of this session is one this version no longer replays — errorCode
+            `checkpoint.unsupported_version` (`retriable: true` = the card is still pending / cancellable, NOT "a retry will succeed"; the stored body carries
+            `settings.permissions.allow` under an earlier reading — list or statement era) or errorCode `resume_blocked_by_policy` (`blockedBy: "request.field_invalid"`;
+            the stored skill `disallowedTools` carries an entry the engine now reads as nothing). Both answers are word-for-word the ones `POST /v1/sessions/{sessionId}/wake`
+            gives for the same body. Way out: cancel the task and resubmit as a new one (with the entry fixed). The SDK does not retry them (409 is not in the retry set).
           content:
             application/json:
               schema: { $ref: '#/components/schemas/ErrorResponse' }
@@ -1022,7 +1031,7 @@ paths:
             Use POST /v1/runs/{taskId}/cancel for a run-level stop.
             errorCode "steering.not_running" (reused verbatim) — terminal / not accepting, or the stream
             settled in the race window ("run just finished on this replica — no in-flight turn to interrupt").
-            errorCode "steering.duplicate_input_id" — this Idempotency-Key was already accepted with DIFFERENT
+            errorCode "steering.duplicate_input_id" (server >= 7.118.0: read `detail.differs` first, see SteerReplayFacet) — this Idempotency-Key was already accepted with DIFFERENT
             content or at a DIFFERENT tier (core's replay identity includes the normalized tier: the same key
             that once went through the steer door (default tier) then hits this door (now) is refused —
             "idempotent success without the cut" would be a dispositive lie). Reissue with a fresh key.
@@ -2526,6 +2535,16 @@ paths:
         '409': { $ref: '#/components/responses/Conflict' }
         '422': { $ref: '#/components/responses/UnprocessableEntity' }
         '501': { $ref: '#/components/responses/NotImplemented' }
+        '503':
+          description: >-
+            server >= 7.118.0 — `state.mcp_reconnect_unsettled`: the time this endpoint waits for the re-dial to settle ran out (the same
+            constant as the `/mcp` panel, 10 s). Read it as "this endpoint did not wait for a result", NOT "failed": the re-dial is NOT
+            cancelled and still settles (success writes the roster onto this run, failure removes it). The body is `error` + `errorCode`
+            only (no `outcome`) and carries NO `Retry-After`; this is a close-then-dial transaction — confirm the server state before
+            re-sending, and do not use a re-send as a probe. SDK → ServiceStateError (never auto-retried: a non-submit POST).
+          content:
+            application/json:
+              schema: { $ref: '#/components/schemas/ErrorResponse' }
 
   /v1/sessions/{sessionId}/background:
     parameters:
@@ -3315,8 +3334,11 @@ paths:
             `approval_binding_mismatch` (`errorCode`, `field`: boundCallId|boundInputHash|answer — the third
             member since core 5.7.0/RB-459: the content-ask `answer` payload was refused) — your binding does
             NOT match the current pending action (TOCTOU/swap). The SDK splits these by `errorCode` into
-            ConflictError vs ApprovalBindingMismatchError. binding_mismatch = SAFETY signal: refetch + re-present
-            to the human, NEVER auto-retry. (S-57, server >= 7.54.0: the `field:"boundCallId"` shape is
+            ConflictError vs ApprovalBindingMismatchError (the engine's own code for the same refusal, `checkpoint.invalid_outcome`,
+            maps to that class too). binding_mismatch = SAFETY signal: refetch + re-present
+            to the human, NEVER auto-retry. A `decidedBy: "policy"` allow on a person-required card is the OTHER `invalid_outcome`
+            shape — the non-binding seat `hostDecision` — and answers `resume_outcome_invalid` (`field: "hostDecision"`, card pending;
+            see ApprovalDecidedBy). (S-57, server >= 7.54.0: the `field:"boundCallId"` shape is
             intercepted in steady state by the `approval_stale` second trigger source below — the server
             compares the echoed boundCallId against the current pending pre-CAS and answers 409
             `approval_stale` + `currentPending`; this code's boundCallId arm covers only the remaining race
@@ -3344,7 +3366,8 @@ paths:
             decide either resumes the parked host or starts that new run (both land on the byte-identical
             acceptance shape) — a host session already busy with another run instead answers the ordinary 409
             `conflict.session_active_run`.
-            EIGHTH group (server >= 7.75.0 / S-252, same PARKED WORKFLOW-CHILD lane): `decide.workflow_parks_unreadable`
+            EIGHTH group (server >= 7.75.0 / S-252, same PARKED WORKFLOW-CHILD lane; the 7.75.0–7.117.x reading — server >= 7.118.0
+            narrows the triggers, see the note after this group): `decide.workflow_parks_unreadable`
             — the workflow run's record carries no readable `parks[]` list (the ENGINE's own park-ownership
             ledger, and the single source this lane joins the caller's checkpoint against). A record written by
             an engine older than 7.17.0 does not carry the key at all, or one entry is damaged and the whole list
@@ -3368,6 +3391,15 @@ paths:
             there is no mapper branch, class, or type-union member to delete or keep for it — unlike
             `decide.workflow_host_not_parked` above, an OPEN decide.* code this SDK never gave a name to degrades
             through the base `DecideError` exactly as it always did, on any server version.
+            🔴 server >= 7.118.0 (core 7.44.0 #1319): routing and redemption read ONE source — the holder seat `parkedBy` the engine writes on
+            a `parks[]` entry (the record naming the card at the highest generation, exactly one ⇒ it takes the decision). The EIGHTH group
+            widens and now has two causes under one code (the body carries no cause word; the sentence differs): an entry written by an older
+            engine (cards still parked at the upgrade) ⇒ resume `runId` and decide again; a record that cannot be read (a malformed entry,
+            or `parks` present but not a list) ⇒ resuming does NOT clear it — an operator must repair that record; the body may add
+            `readableHolderRunId` (a clue). A record with NO `parks` key holds nothing: a card only it names answers 409
+            `conflict.resume_context_unavailable` (no `runId`). NINTH group: `decide.workflow_park_forked` (409, body `runIds`) — the same
+            run was resumed more than once and the copies all hold this card; nothing delivered, the card stays PENDING; resume one of
+            `runIds` and decide again (SDK → `DecideWorkflowParkForkedError`).
             SIXTH group (A-075.10, TIME-BASED retry-later pair): `resume.usage_window_exhausted` (#449 G1,
             core 5.60.1 — the deployment governance window on this run's ledger key is full; NOTHING was
             consumed or unpinned, the SAME token with the SAME decision redeems once the window slides) and
@@ -4225,32 +4257,14 @@ paths:
                     substitution), the command bytes were unreadable, or this ask belongs to one of the seven
                     families that never get offers — so do NOT diagnose the deployment's rule lane from one
                     frame or from that one refusal.
-                settledBy:
-                  type: string
-                  enum: [policy]
+                decidedBy:
+                  allOf: [{ $ref: '#/components/schemas/ApprovalDecidedBy' }]
                   description: >-
-                    server #1016 (>= 7.93.3, core 7.26.2), ADDITIVE — the SETTLEMENT SOURCE of this answer.
-                    `"policy"` is a MACHINE REFUSAL, NEVER THE USER'S WORDS: the deployment's own policy answered
-                    this card, not a person. ABSENT = today's shape (read as a person's answer).
-
-                    Same word, same source and the same three-question rationale as the durable leg's key
-                    (`POST /v1/approvals/{sessionId}/decide`, ApprovalDecision.settledBy) — see that schema for
-                    the full account. `decision:"deny"` ONLY: an allow-family decision carrying this key is a
-                    contradiction, refused BEFORE THE DECISION (400 `request.body_shape`, the same sentence as
-                    the durable leg) rather than after — the engine's refusal cannot travel back up this HTTP
-                    leg, and by then this door's own side effects (the session allow-all grant, a persisted
-                    rule, a DECIDED row that replays as a plain approval) have already landed. Because the
-                    refusal precedes the decision, the ask is NOT settled: the card is still open and the same
-                    `approvalId` can be answered again. The PARKED late-acceptance arm reads the same key.
-
-                    SHAPE, not word: a non-string is a 400. A word core does NOT know travels UNTOUCHED — this
-                    door still answers 200 (its 200 only ever meant "your answer was delivered") and the ENGINE
-                    refuses it fail-closed, settling as `approver_contract`, NEVER as `policy_refused`: a
-                    misspelled word must not be rendered as "the deployment's policy refused".
-
-                    No capability bit: an older server ignores unknown keys and answers 200, so the refusal is
-                    still read as a person's there. Judge by server version, or read back whether
-                    `tool_end.gate.settlement.kind` ever shows `policy_refused`.
+                    server >= 7.118.0 — WHO made this decision (one key read the same way by all three decision doors; see
+                    ApprovalDecidedBy). ABSENT = `person`. This door's own answers: a `policy` allow on a person-required card
+                    and a word outside the closed set are both 400 `request.body_shape`, refused BEFORE the decision (the card is
+                    still open; answer it again with the same `approvalId`). 🔴 An older server (no `approvalDecidedBy`
+                    capability bit) DROPS this key silently and answers 200 — probe the bit first.
                 note:
                   type: string
                   maxLength: 2048
@@ -4342,7 +4356,7 @@ paths:
             application/json:
               schema: { $ref: '#/components/schemas/AskDecisionAck' }
         '400':
-          description: 'Malformed body — `request.invalid_json` (unparseable) or `request.body_shape` (schema reject, e.g. an `actor` carrying anything but `label`, or — on a server < 7.104.0 — a `settledBy` key). Server >= 7.104.0 (S-843) adds the `settledBy` refusals shared with the durable `/decide` leg: `request.field_conflict` (with a non-deny decision), `settled_by_not_in_proof` (direct-door worker), `request.field_invalid` (non-string, or outside the 1..16-character column) — see AskDecisionBody.settledBy.'
+          description: 'Malformed body — `request.invalid_json` (unparseable) or `request.body_shape` (schema reject, e.g. an `actor` carrying anything but `label`, or — on a server < 7.104.0 — a `settledBy` key). Server >= 7.118.0 reads `decidedBy` (see AskDecisionBody.decidedBy): `request.field_conflict` (a `policy` allow on a person-required card), `decided_by_not_in_proof` (direct-door worker), `request.field_invalid` (outside the closed set); the retired `settledBy` key is a `request.body_shape`.'
           content:
             application/json:
               schema: { $ref: '#/components/schemas/ErrorResponse' }
@@ -6842,16 +6856,6 @@ components:
               name: { type: string }
               description: { type: string }
               content: { type: string }
-              manifest:
-                type: object
-                additionalProperties: true
-                deprecated: true
-                description: >
-                  RETIRED — server >= 7.116.0 rejects the WHOLE skill when this key is present on a delegated agent's skill (unknown key:
-                  `agent.skills must be …`; a top-level skill answers 400 `request.field_invalid` "skill.manifest was retired in core 7.41.0 — …").
-                  core 7.41.0 retired the skill manifest (hard break, no alias). Rewrite `manifest.allowTools` to the skill's
-                  `allowedTools` (an allow-rule source, not a fence); `manifest.allowPaths` has no replacement (use the session rules'
-                  `allowDirs`). Older servers (<= 7.115.x) still accept it; the key stays declared here until the next major.
               allowedTools: { type: array, items: { type: string }, description: 'server >= 7.116.0 — the skill''s allow-rule source (see `SkillSpec.allowedTools`); <= 7.115.x rejects the key (unknown key).' }
               disallowedTools: { type: array, items: { type: string }, description: 'server >= 7.116.0 — the skill''s deny-rule source (see `SkillSpec.disallowedTools`); <= 7.115.x rejects the key (unknown key).' }
               files:
@@ -7187,6 +7191,13 @@ components:
           description: 'core 5.11.0: extra READ-ONLY absolute host paths — read fence only (classify read-boundary auto-allow + read_file/grep containment); writes to the same dirs are still refused. Same shape/400 rules as additionalDirectories; servers <6.6.0 silently ignore (harmless additive).'
         images:
           type: array
+          description: >
+            Image input: `{ data, mimeType }` (inline base64) or `{ url }`. 🔴 INLINE-ARM ADMISSION (server >= 7.117.0 / core 7.42.0): `mimeType` must be one of `image/png` /
+            `image/jpeg` / `image/gif` / `image/webp` (the API's type word, not a file extension), the decoded bytes must carry the DECLARED type's file signature, and the decoded size must
+            not exceed 5 MiB — otherwise the engine refuses BEFORE the first model call (after the UserPromptSubmit hook; not one model token spent) with 400 `config.image_invalid`, whose
+            `error` starts with `TaskSpec.images: `. Before 7.117.0 such an image was sent as-is to the model provider and the whole first request failed there (no code to branch on).
+            A base64 string longer than 5 242 880 characters (the model API's 5 MB image cap; the server reads the engine's exported constant) is NOT this code: the submit gate refuses it first
+            with 413 `request.payload_too_large` (zero model calls, no submit hook, no run row). The `{ url }` arm is unchanged. The model must accept image input (`ModelInfo.vision`; absent = unknown, send anyway).
           items:
             oneOf:
               - type: object
@@ -7330,9 +7341,20 @@ components:
             the delivery lane, so raising it only helps a model whose declared window is below 200 000 tokens.
             The engine's own twin code `config.skill_listing_budget_invalid` (400) is NOT reachable through this
             submit leg today — it belongs to callers that build a `TaskSpec` directly.
-            🔴 `permissions.allow` IS A STATEMENT, NOT A WHITELIST (server >= 7.115.0 / S-654; probe
-            `capabilities.settings.permissions.allowGrant === true` — an absent key is an older server that reads the bucket as a
-            list, so put nothing in it there). The bucket is compiled by the SAME rule compiler as `deny` / `ask`: a hit is a
+            🔴 `permissions.allow` — THE READING DEPENDS ON THE SERVER GENERATION (probe `capabilities.settings.permissions`; an absent bit = the older generation).
+            server >= 7.117.0 (`allowLayer === true`, core 7.42.0): the WHOLE list goes to the engine's allow layer and a listed call is NOT ASKED — the engine's ordinary and
+            "nobody stated it" questions are cleared, built-in write gates and shell checks included (like the reference implementation's `permissions.allow`); content forms
+            (`Bash(git:*)`, `Read(//dir/**)`) are accepted; entries the engine reads as nothing (`*`, write-path forms such as `Edit(src/**)`, domain forms, names with whitespace) are NOT
+            refused — the run opens with an `engine_notice` `permission.allowed_tool_not_granted`. Any layer's deny / ask rule, operator `APPROVAL_REQUIRE` / `APPROVAL_NEVER_AUTO` /
+            `MANUAL_MODE_SHELL_GATE=always`, write protection, governance, a hook's ask and plan-mode asks still ask (a matching `request` member then shows as `allowRuleShadowed` on the
+            card). The submit gate judges only the shape (array of strings, at most 1000 entries => 400) and the NAMES (the same judge and the same sentence as `deny` / `ask`: a retired name,
+            an unprefixed short name, or a protocol name whose server segment cannot be minted / is missing / is a wildcard (`mcp__*`) is a whole-request 422) — never the content.
+            🔴 THE 422 SENTENCE (server >= 7.117.0): `error` starts with `request settings cannot be applied as written: ` (<= 7.116.x: a different sentence that said tighten-only — all three
+            buckets share this exit, and the old head misled a caller writing a tightening rule and lied about the allow bucket). Do NOT dispatch on the sentence: it carries no code of its own, so
+            `errorCode` is the status-derived coarse code `request.unprocessable`.
+            The skill lists (`skills[].allowedTools`) read exactly like this bucket from 7.117.0 on.
+            Server 7.115.0–7.116.x (`allowGrant` without `allowLayer`): `permissions.allow` IS A STATEMENT, NOT A WHITELIST — an absent `allowGrant` is a server <= 7.114.x that reads the bucket
+            as a list, so put nothing in it there. The bucket is compiled by the SAME rule compiler as `deny` / `ask`: a hit is a
             reasoned allow = "somebody stated it" (it answers an MCP / A2A server-tool call nobody stated; without it an
             ask-class mode raises a card, and with no approval surface the call is refused); a miss states nothing and the call is
             judged as usual — a tool off the list is NOT refused any more (<= 7.114.x refused it). It does NOT skip the question
@@ -7397,6 +7419,14 @@ components:
           items: { $ref: '#/components/schemas/TaskAgentDefinition' }
           description: Per-task sub-agent definitions (roster additions scoped to this task only).
         interactiveTools: { type: boolean, description: Mount interactive (prompting) tools. }
+        approvalReach:
+          allOf: [{ $ref: '#/components/schemas/ApprovalReach' }]
+          description: >-
+            server >= 7.118.0 (core 7.43.0 #1354) — declares that NOBODY on this run can answer approvals. Passed verbatim to the
+            engine. The live approver callback is no longer this run's approver (`wiring_manifest` `ask.form: "declared_none"`); on a
+            deployment with a parking facility the first gated call PARKS the run (`paused`) — for "refuse and keep going" use
+            `permissionMode: "dontAsk"`. It never approves or releases anything. A value outside the closed set ⇒ 400
+            `request.field_invalid`. Probe `capabilities.approvalReach` first.
         oneShot:
           type: boolean
           description: >
@@ -8901,40 +8931,14 @@ components:
             not inside the HMAC payload, so accepting it unsigned would be tamperable); 501 without an exemption
             store. On the parked-background-agent leg the grant lands on the child's ROOT/HOST session (the whole
             session tree shares it). The 200 body echoes `rememberApplied` when a store is wired.
-        settledBy:
-          type: string
-          enum: [policy]
+        decidedBy:
+          allOf: [{ $ref: '#/components/schemas/ApprovalDecidedBy' }]
           description: >-
-            server #1016 (>= 7.93.3, core 7.26.2), ADDITIVE — WHO made this decision. `"policy"` is a MACHINE
-            REFUSAL, NEVER THE USER''S WORDS: the deployment''s own policy answered this card (a headless lane —
-            CI, a background batch, a shell refusing by a standing deployment posture). ABSENT = today''s shape
-            (`person`), so an unrevised caller is byte-for-byte unchanged.
-
-            WHY IT EXISTS: without the word, a policy refusal is byte-identical to "a person pressed No", and the
-            model reads "the user doesn''t want to proceed; STOP and wait for the user to tell you how to proceed"
-            — a sentence about a person, in a run where none is coming. With it the engine mints a POLICY sentence
-            instead, this request''s `reason` travels as the POLICY''s words (not as the user''s guidance), and the
-            settlement is recorded as `policy_refused` (`who.party: "none"`).
-
-            CONSTRAINTS: `decision:"deny"` only — only a person approves an action that EXECUTES, so
-            `{decision:"approve", settledBy:"policy"}` is a contradiction, refused by the SERVER before the
-            decision (400 `request.field_conflict`, the SAME SENTENCE as the live door) with the engine''s own
-            pre-CAS refusal (`checkpoint.invalid_outcome`) behind it as depth; the server does NOT silently drop
-            the key. It is therefore MUTUALLY EXCLUSIVE with `remember` (which is approve-only) — one body
-            cannot carry both. The vocabulary''s owner is still core: the server accepts the KEY and relays the
-            VALUE verbatim, so a word core does not know is refused by the engine, never by a server-side
-            second judge of the vocabulary.
-            REFUSED on a direct-door worker (400 `settled_by_not_in_proof` — not inside the HMAC payload, same
-            rule as `remember` / `updatedInput` / `checkpointToken`); a non-direct-door deployment is unchanged.
-
-            SHAPE, not word: a non-string is 400 `request.field_invalid` (same family as `reason must be a
-            string`). A word core does NOT know travels UNTOUCHED and is refused BY THE ENGINE pre-CAS:
-            409 + `errorCode: "resume_outcome_invalid"` + `field: "hostDecision"` + `retriable: true`, and the
-            card STAYS PENDING (re-decide the same coordinates with a corrected body). NOTE this family no
-            longer answers `approval_binding_mismatch`: from server 7.93.3 the `policy_ask` gate splits
-            `invalid_outcome` by `field`, and the non-binding seats (`hostDecision`, `approver`) answer
-            `resume_outcome_invalid` — "you approved the wrong action" is a sentence about a different fault.
-            The binding seats and the field-absent shape are byte-for-byte unchanged.
+            server >= 7.118.0 — WHO made this decision (one key read the same way by all three decision doors; see
+            ApprovalDecidedBy). ABSENT = `person`. This door's own answers: a `policy` allow on a person-required card is
+            refused by the ENGINE before the CAS (409 `resume_outcome_invalid` + `field: "hostDecision"`, the card stays
+            pending); outside the closed set ⇒ 400 `request.field_invalid`; any value on a direct-door worker ⇒ 400
+            `decided_by_not_in_proof`. An older server (no `approvalDecidedBy` bit) refuses the key (400 `request.body_shape`).
 
     ApprovalRow:
       type: object
@@ -9339,6 +9343,17 @@ components:
             `noteRecorded` (a live-card-only deployment has no row to write, and a flapping store cannot write
             one; both answer `false` while the decision itself still stands). This bit only says "this build
             knows the key" — same family as `permissionRulesRevoke`, deliberately gated on no facility.
+        approvalDecidedBy:
+          type: boolean
+          description: >-
+            server >= 7.118.0, hardcoded true (a version bit): the three approval decision doors read `decidedBy` (ApprovalDecidedBy).
+            Send it only when this is present; absent = an older server (the durable doors and `plan_review` 400 it, the live
+            `respond` drops it silently and answers 200).
+        approvalReach:
+          type: boolean
+          description: >-
+            server >= 7.118.0, hardcoded true (same word as the request key): TaskRequest.approvalReach is understood. Absent = an
+            older server — do not send the key.
         respondFreeFormRules:
           type: boolean
           description: >
@@ -9414,22 +9429,28 @@ components:
         settings:
           type: object
           description: >
-            🔴 server >= 7.115.0 (S-654), ALWAYS present: `{ permissions: { allowGrant: true } }` — the VERSION bit for "which reading does this
-            server give `settings.permissions.allow` on a task submit". `permissions.allowGrant === true` = the `allow` bucket is compiled into
-            ALLOW RULES: a hit gives "somebody stated it" (so an MCP / A2A server-tool call nobody stated no longer raises a card) — it is NOT
-            "skip the question" and NOT a whitelist (a tool off the list is no longer refused; the engine's own write gate / shell checks still ask
-            on built-in tools, deny / ask still win; a bad-shape or content-form name is a whole-request 422).
-            🔴 ABSENT KEY = an older server (<= 7.114.x), which read `allow` as a WHITELIST (everything off the list refused) — do not put anything
-            in the allow bucket there. NOT the same bit as `taskSettings.permissions` (unchanged, still a boolean: "this server accepts the
-            `settings.permissions` face"). A version bit: constant `true`, never varies with the deployment; the deployment-level "who stated what"
-            (operator env `APPROVAL_GRANT`) is on the operator diagnostics face (`serverGates.approvalGrant`), not here.
+            🔴 server >= 7.115.0, ALWAYS present: `{ permissions: { allowGrant: true } }`; server >= 7.117.0 adds `allowLayer: true` to `permissions` — the VERSION bits for
+            "which reading does this server give `settings.permissions.allow` on a task submit" (both constant `true`, never vary with the deployment: the single-user
+            and `REQUIRE_PRINCIPAL=true` forms carry the same value). Read three levels: `settings` -> `permissions` -> <bit>; an absent / non-boolean bit = a server without it.
+            `permissions.allowGrant === true` (>= 7.115.0) = the `allow` bucket is read as allow RULES, not as a list (a tool off the list is no longer refused).
+            `permissions.allowLayer === true` (>= 7.117.0, core 7.42.0) = the server hands the WHOLE `allow` list to the engine's allow layer: a listed call is NOT ASKED (the engine's
+            ordinary and "nobody stated it" questions are cleared — built-in write gates / shell checks included, like the reference implementation's `permissions.allow`); content forms
+            (`Bash(git:*)`, `Read(//dir/**)`) are accepted; entries the engine reads as nothing (`*`, write-path / domain forms, names with whitespace) are NOT refused — the run opens with an
+            `engine_notice` `permission.allowed_tool_not_granted`; any layer's deny / ask rule, operator `APPROVAL_REQUIRE` / `APPROVAL_NEVER_AUTO` / shell floor / write protection /
+            governance, a hook's ask and plan-mode asks still ask (a matching member then shows as `allowRuleShadowed` on the card). The submit gate judges only the shape and the NAMES (retired
+            name / unprefixed short name / unmintable protocol server segment => 422, at most 1000 entries => 400) — never the content.
+            🔴 ABSENT `allowLayer` with `allowGrant` present = a 7.115.0–7.116.x server: allow is a STATEMENT (it does not clear the engine's own write gate / shell checks; a content form
+            is a whole-request 422). ABSENT KEY altogether (no `settings`) = an older server (<= 7.114.x), which read `allow` as a WHITELIST (everything off the list refused) — do not put
+            anything in the allow bucket there. NOT the same bit as `taskSettings.permissions` (unchanged, still a boolean: "this server accepts the `settings.permissions` face"). The
+            deployment-level "who stated what" (operator env `APPROVAL_GRANT`) is on the operator diagnostics face (`serverGates.approvalGrant`), not here.
           additionalProperties: true
           properties:
             permissions:
               type: object
               additionalProperties: true
               properties:
-                allowGrant: { type: boolean, description: 'Constant `true` on server >= 7.115.0: `settings.permissions.allow` is read as allow RULES (a statement), not as a list. The whole `settings` key being absent = an older server.' }
+                allowGrant: { type: boolean, description: 'Constant `true` on server >= 7.115.0: `settings.permissions.allow` is read as allow RULES, not as a list. The whole `settings` key being absent = an older server.' }
+                allowLayer: { type: boolean, description: 'Constant `true` on server >= 7.117.0 (core 7.42.0): the whole `settings.permissions.allow` list goes to the engine allow layer (listed calls are not asked; content forms accepted; entries read as nothing get an `engine_notice` `permission.allowed_tool_not_granted`). Absent = a server without it (7.115.0–7.116.x: allow is a statement; older: a list).' }
         memoryWrite:
           type: boolean
           description: >
@@ -10014,8 +10035,9 @@ components:
         🔴 server >= 7.116.0 (core 7.41.0): a skill carries two TOOL LISTS, `allowedTools` / `disallowedTools` — the
         run-scoped ALLOW-RULE and DENY-RULE source of the skill once it is activated. NOT a fence (<= 7.115.x the engine read
         the skill's tool list as "only these"): unlisted tools are judged as usual, listed ones are not asked about for the rest
-        of the run — which makes it WIDER than `settings.permissions.allow` (that one stays a statement and does not
-        release the engine's own write gate / shell ordinary asks on built-in tools). Probe `capabilities.skillToolLists`
+        of the run. On server 7.116.x that made it WIDER than `settings.permissions.allow` (then a statement that did not
+        release the engine's own write gate / shell ordinary asks on built-in tools); server >= 7.117.0 reads the two request-level
+        allow sources the SAME way (see `capabilities.settings.permissions.allowLayer`). Probe `capabilities.skillToolLists`
         (absent = an old server that accepts the keys on a top-level skill and silently ignores them). The skill's `manifest`
         key is RETIRED: any value on a top-level skill is a 400 `request.field_invalid` ("skill.manifest was retired in core
         7.41.0 — …"; rewrite `manifest.allowTools` to `allowedTools`; `allowPaths` has no replacement), and on a delegated
@@ -10032,21 +10054,28 @@ components:
             server >= 7.116.0. The skill's run-scoped ALLOW-RULE source (permission-rule spellings: a tool name,
             `mcp__<server>` / `mcp__<server>__*`; `*`, write paths and domain forms grant nothing). Once the skill is activated
             (named in top-level `userInvokedSkills` — no `Skill` card — or activated by the model and the card approved) the
-            listed tools are not asked about for the rest of the run: it clears the engine's ordinary and "nobody stated it" asks,
-            so it is WIDER than `settings.permissions.allow`; deny / ask rules, `APPROVAL_REQUIRE`, `APPROVAL_NEVER_AUTO`, the shell
-            floor, write protection and plan-mode asks are not cleared. NOT a fence: unlisted tools are judged as usual. A non-string-array
-            is a 400 `request.field_invalid`; a protocol name whose server segment cannot be minted / is missing / is a wildcard
-            (`mcp__*`) is a 400 too (top-level `skills[]` and `agents[].skills[]` judge identically). Probe `capabilities.skillToolLists`.
+            listed tools are not asked about for the rest of the run: it clears the engine's ordinary and "nobody stated it" asks;
+            deny / ask rules, `APPROVAL_REQUIRE`, `APPROVAL_NEVER_AUTO`, the shell floor, write protection and plan-mode asks are not cleared.
+            NOT a fence: unlisted tools are judged as usual. Versus `settings.permissions.allow`: on server 7.116.x this list was WIDER (that bucket
+            was then only a statement); server >= 7.117.0 reads the two the SAME way, entry for entry (entries the engine reads as nothing get an
+            `engine_notice` `permission.allowed_tool_not_granted`). A non-string-array is a 400 `request.field_invalid`; a protocol name whose server
+            segment cannot be minted / is missing / is a wildcard (`mcp__*`) is a 400 too, and from 7.117.0 so is a retired name / unprefixed short
+            name (the name layer is judged by the same judge as the allow bucket) and a list of more than 1000 entries (top-level `skills[]` and
+            `agents[].skills[]` judge identically). Probe `capabilities.skillToolLists`.
         disallowedTools:
           type: array
           items: { type: string }
           description: >
             server >= 7.116.0. The skill's run-scoped DENY-RULE source (same grammar / gates / probe as `allowedTools`): once the skill
             is activated these tools are refused for the run (no execution, no card; `tool_end.gate.disposition` is
-            `{ kind: "denied", deniedBy: "policy" }`). `*` and domain forms (`WebFetch(domain:…)`) read as "none" on this side too —
-            accepted with 2xx and only an operator-audience notice `skills.allowed_tool_not_granted`; write-path forms and the rest compile
-            into deny rows (as the server contract states it for the engine's `compileSkillGrant`; the SDK did not re-run the engine on the deny side's write-path forms —
-            the engine's own notice text lists "a write-path or domain entry" among entries that "grant (or refuse) nothing").
+            `{ kind: "denied", deniedBy: "policy" }`); write-path forms and command rules compile into deny rows too.
+            🔴 server >= 7.117.0: an entry the engine reads as nothing (`*`, a domain form `WebFetch(domain:…)`, a protocol name with a wildcard, a name with
+            whitespace or malformed) is refused AT SUBMIT — 400 `request.field_invalid`, `error` = `skill.disallowedTools entry "<entry>" reads as nothing (<reason>): <how to write it>`
+            (delegated level: prefix `agents[<i>]: skills[<j>]: `), top-level `skills[]` and `agents[].skills[]` alike — because an entry that looks like a refusal but refuses nothing is
+            worse than none (<= 7.116.x accepted it with only an operator-audience notice that was not sent to the session stream; that code is now `permission.allowed_tool_not_granted`,
+            user audience, on the session stream). 🔴 The resume legs judge the stored body the same way: a parked task whose stored skill `disallowedTools` carries such an entry answers
+            409 `resume_blocked_by_policy` (`blockedBy: "request.field_invalid"`, same sentence) BEFORE the claim — row and card untouched; a terminal run's `POST /v1/runs/{taskId}/steer`
+            answers it before minting the wake card. Why: engine 7.41 trimmed whitespace (`"Bash "` was a working deny), engine 7.42 reads verbatim (reads as nothing). The allow side does not refuse.
 
     McpServerSpec:
       type: object
@@ -10788,16 +10817,17 @@ components:
         baseUrl/apiKey/headers (the service strips them).
       # 🔴 封闭 + required 补铸造点(census 批2 第三段,2026-07-30):铸造点 `routes/capabilities.ts`
       # 是逐键字面量投影(白名单,不是透传——「open set」旧注不成立,新键要先过这道 spec)。core Model 类型
-      # 里 id/provider/reasoning 都是必填、`vision` 由 `input` 恒算 ⇒ 五键无条件;contextWindow/maxOutputTokens
-      # 0/缺省即省略,supportedEffortLevels 仅 reasoning 模型。
-      required: [name, id, provider, reasoning, vision]
+      # 里 id/provider/reasoning 都是必填 ⇒ 四键无条件;contextWindow/maxOutputTokens 0/缺省即省略,
+      # supportedEffortLevels 仅 reasoning 模型。🔴 server 7.117.0 起 `vision` 是三态(收 / 不收 / 没声明 `input`),
+      # 没声明 ⇒ 键缺席 ⇒ 不再是必有键(此前由 `input` 恒算成布尔,对没声明的条目报 false)。
+      required: [name, id, provider, reasoning]
       additionalProperties: false
       properties:
         name: { type: string, description: 'the @mention handle (user-pickable model name).' }
         id: { type: string, description: 'underlying model id (display, e.g. deepseek-v4-pro).' }
         provider: { type: string }
         reasoning: { type: boolean, description: 'a reasoning-tier model (picker can badge it).' }
-        vision: { type: boolean, description: 'accepts image input.' }
+        vision: { type: boolean, description: 'Whether this catalog row accepts image input — THREE states (server >= 7.117.0; before it a row that declared no `input` reported `false`): `true` = the row declares `input` and it contains "image"; `false` = declares `input` without it; ABSENT = the row declared no `input` (a hand-written catalog entry; the env lane and the config-center lane always declare it). ABSENT means UNKNOWN, not false: treat it as "send the image anyway and tell the user", do not read it as `false` and block the image locally — the server hands the image to the model for that same row at run time. Only an explicit `false` justifies "this model does not take images".' }
         contextWindow: { type: integer, description: 'E4 (SHIPPED) — context-window token DENOMINATOR for the %-used gauge; omitted when unknown.' }
         maxOutputTokens: { type: integer, description: 'E4 (SHIPPED) — single-call max output tokens; omitted when unknown.' }
         supportedEffortLevels:
@@ -12260,6 +12290,16 @@ components:
             toolName: { type: string }
             summary: { type: string }
             touchedPaths: {}
+            allowRuleShadowed:
+              type: array
+              items: { $ref: '#/components/schemas/AllowRuleShadow' }
+              description: >
+                server >= 7.117.0 (core 7.42.0): the DURABLE dual of `ToolApprovalFrame.allowRuleShadowed` — the allow-layer members that matched this call
+                and could not clear its ask (same shape and semantics; entries are secret-redacted again at the server's read boundary). Passed through whole
+                on `GET /v1/approvals` and `/v1/approvals/stream`. Absent != "you have no rule". 🔴 The old key `shadowedRule` (a string, server <= 7.116.x)
+                is RETIRED: the SQL backend rewrote its parked rows once at the 7.117.0 boot to `[{ member: "persisted", entries: [<old value>] }]`, but the
+                file backend's pre-upgrade parked rows and run results already on disk (`checkpointGate.riskDescriptor`) keep the old key verbatim
+                (text still redacted, no structured disclosure) — a consumer reading those must still tolerate the old string.
         # server 7.16.0 [3684]② — durable 队列行归因两键(与活卡 tool_approval 帧的同名键语义同族、判据不同源;
         # 两条腿对同一只 ask 的在场性可以不一致,两侧都只认在场、都禁把缺席读成否定)。
         ruleSuggestions:
@@ -12479,6 +12519,13 @@ components:
             the vocabulary's owner is still core and it has grown before, so read it the same way as everywhere
             else: branch known words, keep a `default` arm, and treat an unknown one as "unknown origin", never
             as "no origin". A row minted before the bit existed reads absent (unreported), not "no authority".
+        boundGrantDigest: { type: string, description: 'server >= 7.118.0 — a parked `Skill` card''s grant-binding digest; same value as the live frame''s ToolApprovalFrame.boundGrantDigest.' }
+        boundGrants:
+          allOf: [{ $ref: '#/components/schemas/BoundSkillGrants' }]
+          description: 'server >= 7.118.0 — the value `boundGrantDigest` hashes, published ONLY when it hashes to it (the engine judges; a pair that disagrees reads as absent). DEPLOYMENT TEXT: render with contextual escaping.'
+        classifierNoRuling:
+          allOf: [{ $ref: '#/components/schemas/AutoModeFallbackCause' }]
+          description: 'server >= 7.118.0 — only on a parked `classifier_no_ruling` card: why the auto-mode classifier did not rule.'
         objective: { type: ['string', 'null'], description: 'task objective, null if unavailable.' }
         input:
           description: >
@@ -12740,7 +12787,11 @@ components:
             them is not known" narrowed to this; it still does NOT say "never consumed"; see `engine_notice`
             `task.user_steer_undrained{cause:"crashed"}`). Converged rows only, one row per (taskId, messageId),
             newest 500. From server >=7.111.0 (S-966) the rows include SUBAGENT steers (recorded under the run on the
-            receipt; same row shape, `kind` is still only `steer`; `messageId` = the subagent steer receipt's).
+            receipt; same row shape; `messageId` = the subagent steer receipt's).
+            `kind` = the engine LANE the input was queued on at acceptance (server >= 7.118.0 / core 7.44.0 #1362, closed two words):
+            `steer` (priority `now` / `next` / absent) or `follow_up` (priority `later`, queued for the run's end); the crash notices
+            split the same way (`task.user_steer_undrained` / `task.user_followup_undrained`, the latter with
+            `detail: { followUp, cause: "crashed", taskId, sessionId }`). Read both words; branch on `cause`.
             ABSENT = this read has nothing to report (zero rows, or the ledger could not be read —
             the authoritative notice is the frame on the run stream); the `?message=` single-entry form never
             carries it.
@@ -12750,7 +12801,7 @@ components:
             additionalProperties: false
             properties:
               messageId: { type: string }
-              kind: { type: string, enum: [steer] }
+              kind: { type: string, enum: [steer, follow_up] }
               taskId: { type: string, description: The run the input was accepted for (its 200/202 receipt). }
               acceptedAt: { type: string, description: ISO timestamp. }
               cause: { type: string, enum: [crashed] }
@@ -13550,6 +13601,20 @@ components:
             🔴 NOT interchangeable with `taskId`: the workflow-child park lane mints no taskId at all (this
             decision never landed on any run, so minting one would be a lie), and the other two decide lanes
             mint no runId. Branch by `errorCode`, never fall back from one handle to the other.
+        readableHolderRunId:
+          type: string
+          description: >-
+            server >= 7.118.0, on `decide.workflow_parks_unreadable` (409) ONLY and only on its "the record cannot be read" cause: the
+            readable record that holds the card at the highest generation (absent when there is none or a tie — never null). Where the
+            work can be carried on — a CLUE, not a way to unblock this decision (the damaged record `runId` names keeps it undeliverable
+            until it is repaired). SDK reads it as `DecideWorkflowParksUnreadableError.readableHolderRunId`.
+        runIds:
+          type: array
+          items: { type: string }
+          description: >-
+            server >= 7.118.0, on `decide.workflow_park_forked` (409) ONLY: the resumed copies of one workflow run that all hold this card at
+            the same generation (>= 2). Recovery = resume ONE of them, then decide again; retrying the decide never succeeds.
+            SDK reads it as `DecideWorkflowParkForkedError.runIds`.
         cause:
           type: string
           description: >-
@@ -13628,6 +13693,8 @@ components:
             `not_found.resource` / `request.unprocessable` / `capability.unavailable` …). ALWAYS PRESENT on server >= 7.104.0
             (S-607); before that it was absent when the original refusal carried no code. SDK:
             `ResumeBlockedByPolicyError.blockedBy` (`undefined` when the key is absent or not a string — never defaulted).
+            server >= 7.117.0 adds a third fold source: a stored skill `disallowedTools` entry the engine now reads as nothing (the submit leg's 400 `request.field_invalid` sentence)
+            ⇒ `blockedBy: "request.field_invalid"`, refused before the claim; the terminal-run steer leg answers it before minting the wake card.
 
     ResumeLegConflictError:
       description: >
@@ -14204,11 +14271,14 @@ components:
         unattended leg run under `UNATTENDED_PERMISSION_MODE=dontAsk`).
     AskClass:
       type: string
-      enum: [content_question, hook, ancestor, unverifiable_delete, transcript_write, org_rule, governance_unavailable, rule_unreadable, ask_rule, denial_limit_fallback, requires_user_interaction, operator_forced, shell_floor, write_protection_unresolved, probe, write_protection, tool_marks_mcp, tool_marks_a2a, policy_rra, artifact_shared_read, mode, unstated, ordinary]
+      enum: [content_question, hook, ancestor, unverifiable_delete, transcript_write, org_rule, governance_unavailable, rule_unreadable, ask_rule, denial_limit_fallback, classifier_no_ruling, requires_user_interaction, operator_forced, shell_floor, write_protection_unresolved, probe, write_protection, tool_marks_mcp, tool_marks_a2a, policy_rra, artifact_shared_read, mode, unstated, ordinary]
       description: >
-        core `ASK_CLASSES` (twenty-two words as of core 7.40.0, twenty-three as of core 7.41.0 — 7.34.0 minted twenty; `shell_floor` joined in 7.39.0 #1260 and rides
+        core `ASK_CLASSES` (twenty-two words as of core 7.40.0, twenty-three as of core 7.41.0, twenty-four as of core 7.43.0 — 7.34.0 minted twenty; `shell_floor` joined in 7.39.0 #1260 and rides
         server >= 7.113.0; `unstated` joined in 7.40.0 and rides server >= 7.115.0; `requires_user_interaction` joined in 7.41.0 and rides
-        server >= 7.116.0) — which CLASS an ask falls in (the columns of the mode × class arbitration table).
+        server >= 7.116.0; `classifier_no_ruling` joined in 7.43.0 #1353 and rides server >= 7.118.0) — which CLASS an ask falls in (the columns of the mode × class arbitration table).
+        `classifier_no_ruling` = the auto-mode classifier did not rule (it could not run — `error` / `timeout` — or replied with no verdict —
+        `parse_error`), so a person decides: the denial-limit fallback's row (bypass-immune, no allow rule clears it, the classifier is not asked
+        again); the live frame and the inbox row also carry the cause as `classifierNoRuling`.
         `requires_user_interaction` = the MCP server declared, in the tool's advertisement, `_meta["anthropic/requiresUserInteraction"] === true`
         (only the boolean `true`): EVERY call of that tool asks the human, in EVERY mode including `bypassPermissions` (it is not one of the ten
         RELEASABLE classes), and only the human's answer clears it — `settings.permissions.allow`, operator `APPROVAL_GRANT`, a skill's tool list +
@@ -15905,29 +15975,39 @@ components:
             requiresRealApproval / shadowed / hook-produced / inherited-unresolved / ancestor-resolved /
             anonymous asks. Absence reads as "THIS CARD has no such option", never as "this deployment has no
             rule lane" (that is `capabilities.permissionRules`) and never as "your rule broke" (that is
-            `persistedRuleShadowed`).
+            `allowRuleShadowed`; `persistedRuleShadowed` on a server <= 7.116.x).
 
             ⚠️ REDEMPTION SCOPE is the live respond leg keyed by this frame's `approvalId` only; a replayed
             offer on another replica is not a promise that it is still redeemable.
         persistedRuleShadowed:
           type: string
-          description: >
-            server >= 7.13.0 (core 5.25.0 #144), ADDITIVE, "tool_approval" only: this ask MATCHED one of the
-            caller's persisted allow rules and that rule COULD NOT silence it. The value is the shadowed rule's
-            VERBATIM text (secret-redacted + length-bounded). UNTRUSTED for display, same treatment as `message`.
-
-            It documents a BEHAVIOUR NARROWING ("Allow rules silence the CLASSIFIER's questions, never a MANDATED
-            one"): under MANUAL_MODE_SHELL_GATE=always, for tools carrying an egress/irreversible mark, and under
-            the governance gates, asks a rule used to silence now REAPPEAR. Render an explanatory line ("your rule
-            is still there — this call must be confirmed each time"); without it the only conclusion a user can
-            draw is "my rule broke", while the rule sits fine in the store.
-
-            🔴 ABSENT != "you have no rule": present only when a rule matched AND could not clear the ask. Most
-            asks have no match at all, and matched-and-cleared ones never become asks (they were silenced — no
-            card exists). 🔴 Deliberately SEPARATE from `governanceForced` and not derivable from it: that key
-            answers "who imposed the gate", this one answers "what happened to YOUR rule"; governance is only one
-            of three un-silenceable sources (doctrine `always` and a tool's own mark carry no governance badge).
-            The durable dual is `gate.riskDescriptor.shadowedRule` on the parked row.
+          description: >-
+            server >= 7.13.0 (core 5.25.0 #144), ADDITIVE, "tool_approval" only — this ask MATCHED one of the caller's persisted allow
+            rules and that rule COULD NOT silence it; the value is the rule's VERBATIM text (secret-redacted + length-bounded; UNTRUSTED
+            for display, like `message`). Absent != "you have no rule"; deliberately separate from `governanceForced`.
+            RETIRED: server >= 7.117.0 (core 7.42.0, BREAKING rename, no alias) never mints this key on any face; its successor is
+            `allowRuleShadowed` (all five allow-layer members, not just the rule store). Kept for servers in [7.13.0, 7.117.0) only (sdk 16.3.0).
+        allowRuleShadowed:
+          type: array
+          items: { $ref: '#/components/schemas/AllowRuleShadow' }
+          description: >
+            server >= 7.117.0 (core 7.42.0), ADDITIVE, "tool_approval" only — BREAKING RENAME, replacing the retired string `persistedRuleShadowed`
+            (no alias): the allow-layer members that MATCHED this call and could NOT clear this ask, one `{ member, entries, omitted? }` record per
+            matching member (see AllowRuleShadow). `member` is one of `hook` / `deployment` / `persisted` / `skill` / `request`; `entries` are the
+            entries that member matched, as their source spells them (UNTRUSTED for display like `message`, secret-redacted, <= 512 chars each,
+            <= 32 listed; a `hook` record's `entries` is always the empty list); `omitted` counts the entries not listed.
+
+            The explaining sentence ("... matches this call but does not clear the ask — <reason>") is IN `message` (the engine writes it where it builds
+            the card); this key is the structured twin of the same fact. Render a line from it ("your rule matched, but this call must be confirmed each
+            time"): without it the only conclusion a user can draw is "my rule broke".
+
+            🔴 ABSENT != "you have no rule": present only when a member matched AND could not clear the ask; a match that cleared never becomes an ask.
+            🔴 `member: "hook"` does not say whose hook (a deployment PreToolUse hook and, on the single-user lane, the request's own `settings.hooks`
+            share one member). 🔴 Deliberately SEPARATE from `governanceForced` (that one: "who imposed the gate"; this one: "what happened to YOUR
+            rule"). What the allow layer can NOT clear: any layer's deny / ask rule, operator `APPROVAL_REQUIRE` / `APPROVAL_NEVER_AUTO` /
+            `MANUAL_MODE_SHELL_GATE=always`, write protection, sensitive-path writes, governance, a hook's ask, plan-mode asks. The durable dual is
+            `gate.riskDescriptor.allowRuleShadowed` on the parked row (PendingCheckpoint.riskDescriptor). Rolling-upgrade caveat: a card stored by a
+            <= 7.116.x replica (`card_json` carrying `persistedRuleShadowed`) cannot be read by a >= 7.117.0 replica — the replay skips it.
         probeCause:
           type: object
           description: >
@@ -16124,6 +16204,24 @@ components:
             `ASK_CLASSES` order, non-empty, `classes[0] === askClass`); one ask can belong to several, so read the whole
             set, not just `askClass`. Present together with `askClass`; passed through verbatim in order (the server
             neither sorts nor dedupes).
+        boundGrantDigest:
+          type: string
+          description: >-
+            server >= 7.118.0 (core 7.43.0 #1348 / #1357), "tool_approval" only — on a call of the engine's `Skill` tool: the hash of what
+            approving it ALSO grants (the skill name + its two lists, canonical spelling); absent on every other card. Same value as the
+            parked row's / the inbox row's twin (the live card and its parked twin reconcile by `(toolCallId, boundGrantDigest)`); passed
+            through verbatim. A live "allow for this session" is remembered under "this tool + this digest" and answers only an ask with the
+            same digest (same skill, same lists); a different skill or a changed list asks again. The durable `remember: "session"` never
+            remembers a skill card.
+        boundGrants:
+          allOf: [{ $ref: '#/components/schemas/BoundSkillGrants' }]
+          description: 'server >= 7.118.0, "tool_approval" only — the value `boundGrantDigest` hashes; present iff it is. DEPLOYMENT TEXT: render with contextual escaping.'
+        classifierNoRuling:
+          allOf: [{ $ref: '#/components/schemas/AutoModeFallbackCause' }]
+          description: >-
+            server >= 7.118.0 (core 7.43.0 #1353), "tool_approval" only, only on a `classifier_no_ruling` card — WHY the auto-mode classifier
+            did not rule (`error` / `timeout` = it could not run; `parse_error` = it replied with no verdict), so a person decides. The server
+            screens it with the engine's closed set (an out-of-set word is withheld). Display only.
         denialLimitFallback: { $ref: '#/components/schemas/DenialLimitFallback' }
         parked:
           type: boolean
@@ -16320,12 +16418,19 @@ components:
             promise and the per-leg index rules are stated verbatim on ToolApprovalFrame.ruleOffers.
         persistedRuleShadowed:
           type: string
+          description: >-
+            server >= 7.13.0 (core 5.25.0 #144), ADDITIVE — the persisted rule this ask matched but which could not silence it; frame and
+            card carry the SAME material (see ToolApprovalFrame.persistedRuleShadowed).
+            RETIRED: server >= 7.117.0 (core 7.42.0) never mints this key on any face (its strict card schema even refuses a stored card
+            carrying it); its successor is `allowRuleShadowed`. Kept for servers in [7.13.0, 7.117.0) only (sdk 16.3.0).
+        allowRuleShadowed:
+          type: array
+          items: { $ref: '#/components/schemas/AllowRuleShadow' }
           description: >
-            ADDITIVE (server >= 7.13.0 / core 5.25.0 #144): the VERBATIM text of the persisted rule this ask
-            matched but which could not silence it (secret-redacted + length-bounded). UNTRUSTED for display.
-            Semantics, the ABSENT != "you have no rule" clause and the deliberate separation from
-            `governanceForced` are stated verbatim on ToolApprovalFrame.persistedRuleShadowed — the live frame,
-            the stored `card_json` and the replayed frame carry the SAME material.
+            ADDITIVE (server >= 7.117.0 / core 7.42.0, BREAKING RENAME of the retired `persistedRuleShadowed`): the allow-layer members that matched this
+            call and could not clear this ask. Semantics, the five members, the always-empty `hook` entries, the explaining sentence living in `message`,
+            the ABSENT != "you have no rule" clause and the separation from `governanceForced` are stated verbatim on ToolApprovalFrame.allowRuleShadowed
+            — the live frame, the stored `card_json`, the replayed frame and the `approval_request` card carry the SAME material.
         fromSubagent: { type: boolean, enum: [true], description: 'Present ⇔ the ask comes from a DELEGATED child — the explicit discriminator (a worker cannot forge it).' }
         sourceTaskId: { type: string, description: 'The child''s core session id — row-join value + fallback discriminator.' }
         sourceAgentName: { type: string, description: 'The child agent''s display name, redacted. UNTRUSTED.' }
@@ -16450,6 +16555,37 @@ components:
         `.strict()`). The eleven known words are exported at runtime as the SDK's `ASK_ORIGINS` (core 7.32.1
         `ASK_ORIGINS`, verbatim and in order) — known words, NOT an enforcement table.
 
+    AllowRuleMember:
+      type: string
+      enum: [hook, deployment, persisted, skill, request]
+      description: >
+        The closed five-word vocabulary of `AllowRuleShadow.member` (server >= 7.117.0 / core 7.42.0 `ALLOW_RULE_MEMBERS`, verbatim and in the order the engine reads the members). The SDK exports
+        it at runtime as `ALLOW_RULE_MEMBERS`. Owner = core; read it as an open string with a default arm (core may add a member and the server follows).
+
+    AllowRuleShadow:
+      type: object
+      additionalProperties: false
+      required: [member, entries]
+      description: >
+        One allow-layer DISCLOSURE record (`ToolApprovalFrame.allowRuleShadowed` / `ApprovalCard.allowRuleShadowed` / the durable
+        `PendingCheckpoint.riskDescriptor.allowRuleShadowed`; server >= 7.117.0 / core 7.42.0 `AllowRuleShadow`): one member of the allow layer that
+        MATCHED the call and could NOT clear its ask. At most one record per member, at most five per ask. The vocabulary owner is core; the server
+        drops the whole key on a member outside the five words, so a >= 7.117.0 server only ever sends these five — still read `member` as an open
+        string with a default arm (core may add members and the server follows).
+      properties:
+        member:
+          $ref: '#/components/schemas/AllowRuleMember'
+          description: >
+            Which allow-layer member matched, in the order the engine reads them: `hook` = a PreToolUse hook's explicit `allow` (a deployment hook or,
+            on the single-user lane, the request's own `settings.hooks` — the key cannot say which) · `deployment` = the operator's `APPROVAL_GRANT` ·
+            `persisted` = a person's stored allow rules · `skill` = an activated skill's `allowedTools` · `request` = the request's `settings.permissions.allow`.
+        entries:
+          type: array
+          maxItems: 32
+          items: { type: string, maxLength: 512 }
+          description: 'The entries that member matched, as their source spells them (UNTRUSTED for display, secret-redacted, <= 512 chars each, <= 32 listed). Always the empty list for `hook` — its word is about the call, not about an entry.'
+        omitted: { type: integer, minimum: 1, description: 'How many matched entries were not listed (beyond the cap). Absent = none omitted.' }
+
     RuleOffersAbsence:
       type: string
       enum: [mandated, shadowed, lane_cannot_speak]
@@ -16458,7 +16594,7 @@ components:
         `ApprovalCard.ruleOffersAbsence`; server >= 7.55.0 / core #490 fix ②, S-15). The engine names which
         door is shut: `mandated` = this ask must be confirmed every time (🔴 this arm must NOT point the user
         at writing a rule — a rule would not silence it, and saying so sends them into a loop); `shadowed` = a
-        rule matched but could not clear it (the machine-readable twin of `persistedRuleShadowed`);
+        an allow-layer member matched but could not clear it (the machine-readable twin of `allowRuleShadowed`; `persistedRuleShadowed` on a server <= 7.116.x);
         `lane_cannot_speak` = the engine could mint no coverable form for this command.
 
         Engine-side MUTUALLY EXCLUSIVE with `ruleOffers`. 🔴 ABSENCE IS NOT AN ASSERTION — it covers both
@@ -16525,21 +16661,25 @@ components:
     RuleOfferMatch:
       type: string
       description: >
-        The MATCH FORM of a persisted rule — the server's `PERSISTED_RULE_MATCHES`, verbatim (the vocabulary
+        The MATCH FORM of a persisted rule — the server's `PERSISTED_RULE_MATCHES`, verbatim and in order (the vocabulary
         is OWNED BY CORE; the server only transcribes it and pins set equality, so core adding a member turns
-        the server red at compile time).
+        the server red at compile time). ONE vocabulary for two faces: a card's rule offer (`RuleOffer.match`, the
+        server's card schema validates it with this very enum) and a persisted rule row (`GET /v1/rules`).
 
         `exact` = this one command only (`Bash(git status)`). `prefix` = the historical colon-star
         word-boundary prefix (`Bash(git status:*)`) and the ONLY spelling for a compound prefix form; it stays
         a permanent compatibility read. `wildcard` = the trailing space-star form (`Bash(npm run *)`), the
-        same predicate as `prefix` on a single command body, minted by core's suggestion face since
-        design/382 / #510. `subpath` = the `Read(//abs-dir/**)` directory form, which never holds a command.
-
-        🔴 FOUR MEMBERS, not "the three today's miner emits": this type describes what a consumer may RECEIVE,
-        and the server's wire validator accepts all four. A consumer type narrower than the mint point is a
-        lie — it lets a `switch` claim exhaustiveness the wire does not honour. (The retired `RuleSuggestion`
-        declared only two, which is exactly why a legal `wildcard` candidate used to be dropped as malformed.)
-      enum: [exact, prefix, wildcard, subpath]
+        same predicate as `prefix` on a single command body. `subpath` = the `Read(//abs-dir/**)` directory form,
+        which never holds a command. `path` (server >= 7.67.0) = the path-family pattern grammar for deny/ask/allow
+        rules (`//abs`, `~/`, root-relative, cwd-relative; `*` within a segment, `**` across segments). `name` /
+        `name_prefix` (server >= 7.118.0 / core 7.43.0 #1318) = the `Skill` name grammar: `Skill(<name>)` exact,
+        `Skill(<prefix>:*)` a name prefix with NO word boundary; a skill card's "don't ask again" offer carries
+        `match: "name"`. A replica rolled back to 7.117.x reads a `name` / `name_prefix` row as a damaged row.
+
+        🔴 SEVEN MEMBERS, closed: the server validates the member on both faces. A consumer type narrower than the
+        mint point is a lie — it lets a `switch` claim exhaustiveness the wire does not honour (SDK <= 16 declared
+        four here, so a `path` offer and every 7.118.0 skill-card offer were dropped as malformed).
+      enum: [exact, prefix, wildcard, subpath, path, name, name_prefix]
 
     RuleOfferUncoveredReason:
       type: string
@@ -17115,25 +17255,66 @@ components:
           additionalProperties: false
           properties:
             label: { type: string, maxLength: 256 }
-        settledBy:
-          type: string
-          enum: [policy]
+        decidedBy:
+          allOf: [{ $ref: '#/components/schemas/ApprovalDecidedBy' }]
           description: >-
-            server S-843 (>= 7.104.0), ADDITIVE — the SETTLEMENT SOURCE of this decision. Same word, same source, same
-            sentence as the durable decide leg's key (`POST /v1/approvals/{sessionId}/decide`, ApprovalDecision.settledBy —
-            see that schema for the full account; the server judges this door and `/decide` with one function, and the live
-            respond leg reads the same two refusal sentences under its own `request.body_shape` family):
-            `"policy"` = the deployment's own policy answered this card, not a person; ABSENT = today's shape.
-            `decision:"deny"` only (an approve carrying it ⇒ 400 `request.field_conflict`); REFUSED on a direct-door worker
-            (400 `settled_by_not_in_proof`); a non-string ⇒ 400 `request.field_invalid`. 🔴 An UNKNOWN word is the OPPOSITE of
-            `/decide` here: this door does not judge the word, so it answers 200 and the word is stored with the decision (the
-            card is settled as `deny`; its source cannot be corrected afterwards, and an idempotent replay replays it), and the
-            engine refuses the word fail-closed at settlement — NOT `/decide`'s 409 `resume_outcome_invalid` with the card left
-            pending for a corrected re-decide. THIS leg only: the value lands in a fixed-width
-            `settled_by` column — outside 1..16 characters ⇒ 400 `request.field_invalid` before the CAS; none of the three
-            response bodies echoes it. OLDER SERVERS (< 7.104.0): this body schema is strict, so the key is an unknown key ⇒
-            400 `request.body_shape` and nothing is decided — a loud refusal, not a silent drop. No capability bit; gate on
-            the server version.
+            server >= 7.118.0 — WHO made this decision (one key read the same way by all three decision doors; see
+            ApprovalDecidedBy). ABSENT = `person`. This door's own answers (both refused before the decision, the card does
+            not move): a `policy` allow on a person-required card ⇒ 400 `request.field_conflict`; outside the closed set ⇒
+            400 `request.field_invalid`; any value on a direct-door worker ⇒ 400 `decided_by_not_in_proof`. An older server
+            (no `approvalDecidedBy` bit) has a strict body here ⇒ 400 `request.body_shape`, nothing decided.
+
+    ApprovalDecidedBy:
+      type: string
+      enum: [person, policy]
+      description: >
+        server >= 7.118.0 (`APPROVAL_DECIDED_BY`, verbatim and in order) — WHO made an approval decision, one key on all three decision
+        doors (live `POST /v1/tool-approvals/{approvalId}/respond`, durable `POST /v1/tasks/{taskId}/asks/{askId}/decision`, durable
+        `POST /v1/approvals/{sessionId}/decide`). `person` = a person decided through the approval channel (ABSENT means this);
+        `policy` = this deployment's own policy decided (a rule / posture answering on someone's behalf, a headless lane) — BOTH
+        directions: the engine records `tool_end.gate.settlement.kind` = `policy_allowed` / `policy_refused` (`who.party: "policy"`).
+        The only verb constraint is the engine's own: a person-required card (`requiresRealApproval`) cannot be ALLOWED by a policy.
+        Send it only when `GET /v1/capabilities` carries `approvalDecidedBy: true`. Outside the closed set (including `sla_timeout` —
+        "the wait ran out with nobody answering" is said only by the server's own deadline sweep — and the old `human` / `timeout`
+        words) ⇒ 400 before the decision. It REPLACES the 7.93.3–7.117.x key `settledBy` (deny-only, the same fact), deleted
+        by server 7.118.0 with no alias: sending `settledBy` ⇒ 400 `request.body_shape` on all three doors.
+        `plan_review` does not take it (400 `request.body_shape`).
+
+    ApprovalReach:
+      type: string
+      enum: [none]
+      description: >
+        The closed set of the request key `approvalReach` (core 7.43.0 #1354 `APPROVAL_REACH`, verbatim): `none` — nobody on this run can be
+        asked to approve. There is no positive word ("a person is reachable" is the `interactionPosture: "interactive"` promise).
+
+    AutoModeFallbackCause:
+      type: string
+      enum: [error, timeout, parse_error]
+      description: >
+        core 7.43.0 `AUTO_MODE_FALLBACK_CAUSES`, verbatim and in order — why the auto-mode classifier did not rule: `error` / `timeout` (it could
+        not run), `parse_error` (it replied with no verdict). Carried as `classifierNoRuling` on a `classifier_no_ruling` card (server >= 7.118.0).
+
+    BoundSkillGrants:
+      type: object
+      description: >
+        core 7.43.0 `BoundSkillGrants` (three keys, verbatim) — what approving an engine `Skill` card ALSO grants: `skill` = the loaded skill's
+        name (`null` = the call names no loaded skill; it grants nothing), `allowedTools` = what activation lets be used without asking,
+        `disallowedTools` = what activation refuses for the rest of the run (canonical spelling: de-duplicated, sorted). DEPLOYMENT TEXT.
+      required: [skill, allowedTools, disallowedTools]
+      additionalProperties: false
+      properties:
+        skill: { type: ['string', 'null'] }
+        allowedTools: { type: array, items: { type: string } }
+        disallowedTools: { type: array, items: { type: string } }
+
+    SteerReplayFacet:
+      type: string
+      enum: [text, trusted, priority, actor]
+      description: >
+        core 7.44.0 #1374 `STEER_REPLAY_FACETS`, verbatim and in order — the facets on which a replayed steering input id differs from the
+        accepted one. Carried as `detail.differs` (a non-empty subset, the engine's order) on 409 `steering.duplicate_input_id` (server >= 7.118.0):
+        contains `text` ⇒ a different instruction, reissue with a fresh key; no `text` ⇒ the SAME instruction — do NOT change the key (it
+        would queue twice), replay with the original values.
 
     AskDecisionAck:
       type: object
@@ -18273,7 +18454,7 @@ components:
           type: object
           description: 'Approval seam: wired form, which seat it came from, and (effective half) what it adjudicates to.'
           properties:
-            form: { type: string, enum: [callback, allow, deny, absent] }
+            form: { type: string, enum: [callback, allow, deny, absent, declared_none], description: '`declared_none` (server >= 7.118.0 / core 7.43.0): the run declared nobody on it can be asked (`approvalReach: "none"`), so a callback beside the declaration is not this run''s approver.' }
             provenance: { type: string, enum: [spec, deps] }
             effective: { type: string, enum: [human_reachable, auto_allow, auto_deny, park_only, unresolved] }
         question:
@@ -19651,17 +19832,11 @@ components:
             engine's catalogue, not on this wire. READ IT; do not switch over a closed list. A name the grammar
             does not know is still a 400 on the input faces.
         match:
-          type: string
-          enum: [exact, prefix, wildcard, subpath, path]
-          description: >
-            The match form (core `PersistedRuleMatch`, five members). `exact` = this one command only ·
-            `prefix` = word-boundary prefix (the historical colon-star spelling, and the only spelling for a
-            compound prefix) · `wildcard` = the space-star form, the SAME predicate as `prefix` on a single
-            command body · `subpath` = the directory (read-containment) form, which never admits a command ·
-            `path` = NEW in server >= 7.67.0, the path-family pattern grammar for deny/ask/allow rules (`//abs`,
-            `~/`, root-relative, cwd-relative; `*` within a segment, `**` across segments).
-            SDK 8.7.0 and earlier declared only the first two — a consumer rendering this cell off a closed list
-            must add the other three.
+          allOf: [{ $ref: '#/components/schemas/RuleOfferMatch' }]
+          description: >
+            The match form — the same seven-member vocabulary as a card's rule offer (`RuleOfferMatch`, the server's
+            `PERSISTED_RULE_MATCHES`): `exact` / `prefix` / `wildcard` / `subpath` / `path`, and from server >= 7.118.0
+            `name` / `name_prefix` (a `Skill(<name>)` / `Skill(<prefix>:*)` rule).
         command: { type: string, description: 'What the matcher actually compares against, parsed out of `rule` at mint time and stored so no consumer re-parses. Trust this + `match`, not your own parse of `rule`.' }
         adds:
           type: array