@stigmer/server 3.38.1 → 3.38.3

package/src/domain/workflow/registry/data/model-registry.json

@@ -963,7 +963,8 @@
       "vision" : true,
       "streaming" : true,
       "thinking" : false,
-      "adaptiveThinking" : true
+      "adaptiveThinking" : true,
+      "thinkingRequired" : true
     }
   }, {
     "id" : "claude-haiku-4.5",
@@ -995,8 +996,9 @@
       "toolUse" : true,
       "vision" : true,
       "streaming" : true,
-      "thinking" : false,
-      "adaptiveThinking" : false
+      "thinking" : true,
+      "adaptiveThinking" : false,
+      "thinkingRequired" : false
     }
   }, {
     "id" : "claude-opus-4.5",
@@ -1017,7 +1019,7 @@
       "effectiveAt" : "2026-05-03T00:00:00Z"
     },
     "contextWindowTokens" : 200000,
-    "maxOutputTokens" : 65536,
+    "maxOutputTokens" : 64000,
     "tokenCounterMethod" : "anthropic_native",
     "summarization" : {
       "triggerThreshold" : 180000,
@@ -1029,7 +1031,8 @@
       "vision" : true,
       "streaming" : true,
       "thinking" : true,
-      "adaptiveThinking" : false
+      "adaptiveThinking" : false,
+      "thinkingRequired" : false
     }
   }, {
     "id" : "claude-opus-4.6",
@@ -1050,7 +1053,7 @@
       "effectiveAt" : "2026-05-03T00:00:00Z"
     },
     "contextWindowTokens" : 200000,
-    "maxOutputTokens" : 131072,
+    "maxOutputTokens" : 128000,
     "tokenCounterMethod" : "anthropic_native",
     "summarization" : {
       "triggerThreshold" : 180000,
@@ -1062,7 +1065,8 @@
       "vision" : true,
       "streaming" : true,
       "thinking" : false,
-      "adaptiveThinking" : true
+      "adaptiveThinking" : true,
+      "thinkingRequired" : false
     }
   }, {
     "id" : "claude-opus-4.7",
@@ -1096,7 +1100,8 @@
       "vision" : true,
       "streaming" : true,
       "thinking" : false,
-      "adaptiveThinking" : true
+      "adaptiveThinking" : true,
+      "thinkingRequired" : false
     }
   }, {
     "id" : "claude-opus-4.8",
@@ -1130,7 +1135,8 @@
       "vision" : true,
       "streaming" : true,
       "thinking" : false,
-      "adaptiveThinking" : true
+      "adaptiveThinking" : true,
+      "thinkingRequired" : false
     }
   }, {
     "id" : "claude-sonnet-4.5",
@@ -1151,7 +1157,7 @@
       "effectiveAt" : "2026-05-03T00:00:00Z"
     },
     "contextWindowTokens" : 200000,
-    "maxOutputTokens" : 65536,
+    "maxOutputTokens" : 64000,
     "tokenCounterMethod" : "anthropic_native",
     "summarization" : {
       "triggerThreshold" : 180000,
@@ -1163,7 +1169,8 @@
       "vision" : true,
       "streaming" : true,
       "thinking" : true,
-      "adaptiveThinking" : false
+      "adaptiveThinking" : false,
+      "thinkingRequired" : false
     }
   }, {
     "id" : "claude-sonnet-4.6",
@@ -1195,8 +1202,9 @@
       "toolUse" : true,
       "vision" : true,
       "streaming" : true,
-      "thinking" : true,
-      "adaptiveThinking" : false
+      "thinking" : false,
+      "adaptiveThinking" : true,
+      "thinkingRequired" : false
     }
   }, {
     "id" : "claude-sonnet-5",
@@ -1230,7 +1238,8 @@
       "vision" : true,
       "streaming" : true,
       "thinking" : false,
-      "adaptiveThinking" : true
+      "adaptiveThinking" : true,
+      "thinkingRequired" : false
     }
   } ]
 }

package/src/domain/workflow/registry/model-registry-store.ts

@@ -63,6 +63,19 @@ export const FAST_VARIANT_KEY = "fast";
  */
 export const THINKING_CAPABILITY_KEY = "thinking";
 
+/**
+ * The registry capability key of adaptive thinking, the second form that
+ * makes THINKING_MODE_ENABLED selectable (stigmer/stigmer#1280).
+ */
+export const ADAPTIVE_THINKING_CAPABILITY_KEY = "adaptiveThinking";
+
+/**
+ * The registry capability key naming a model that always thinks and refuses
+ * a request to turn thinking off: an explicit THINKING_MODE_DISABLED is
+ * refused for it at create.
+ */
+export const THINKING_REQUIRED_CAPABILITY_KEY = "thinkingRequired";
+
 export interface ModelRegistryStoreOptions {
   /** Bundled build-time snapshot; MUST pass the sanity gate at boot. */
   bundledDocument: string;
@@ -202,9 +215,9 @@ export class ModelRegistryStore implements ModelCatalogProvider {
    * "thinking") true under the given harness. Capability flags are
    * harness-scoped facts (they describe what works through that serving
    * path), so there is no any-harness form: THINKING_MODE_ENABLED validates
-   * against the cursor harness specifically — the only harness with a
-   * thinking translation in v1 (oss#772) — and native entries declaring the
-   * same capability stay unselectable until a native wire mapping exists.
+   * against the entry of the harness the execution runs on (oss#772,
+   * stigmer/stigmer#1280), because an id with an entry on both harnesses
+   * can declare different capabilities on each.
    */
   hasCapabilityForHarness(
     harness: string,

package/src/index.ts

@@ -293,12 +293,14 @@ export { RequestContext } from "./pipeline/request-context.js";
 export {
   abortedError,
   alreadyExistsError,
+  ERROR_REASON_DOMAIN,
   failedPreconditionError,
   internalError,
   invalidArgumentError,
   notFoundError,
   unavailableError,
 } from "./pipeline/errors.js";
+export type { RefusalReason } from "./pipeline/errors.js";
 // The shared slug derivation (C2 Stage 3): extension-registered resource
 // kinds derive slugs with the SAME generator both editions pin
 // (ApiRequestResourceSlugGenerator parity) — the semantics live exactly

package/src/pipeline/__tests__/blind-not-found.test.ts

@@ -41,9 +41,6 @@ const PENDING: ReadonlyMap<string, number> = new Map([
   ["domain/agentchannel/controller.ts", 1],
   ["domain/agentinstance/controller.ts", 1],
   ["domain/agentshare/controller.ts", 1],
-  ["domain/artifact/controller.ts", 1],
-  ["domain/environment/controller.ts", 1],
-  ["domain/environment/steps.ts", 1],
   ["domain/plugin/controller.ts", 1],
   ["domain/skill/controller.ts", 1],
   ["domain/workflow/controller.ts", 1],

package/src/pipeline/__tests__/errors.test.ts

@@ -0,0 +1,72 @@
+/**
+ * The structured-refusal seam on failedPreconditionError: a reason rides
+ * as a google.rpc.ErrorInfo detail beside the byte-pinned message, and
+ * reaches a client over a real Connect call. The SDK's getErrorReason
+ * reads the detail the same way the round trip below does, so a refusal
+ * this helper builds is one a console can branch on.
+ */
+import { create } from "@bufbuild/protobuf";
+import {
+  Code,
+  ConnectError,
+  createClient,
+  createRouterTransport,
+} from "@connectrpc/connect";
+import { ErrorInfoSchema } from "@stigmer/protos/google/rpc/error_details_pb";
+import { Health, HealthCheckRequestSchema } from "@stigmer/protos/grpc/health/v1/health_pb";
+import { describe, expect, it } from "vitest";
+
+import { ERROR_REASON_DOMAIN, failedPreconditionError } from "../errors.js";
+
+describe("failedPreconditionError", () => {
+  it("carries no detail without a reason", () => {
+    const error = failedPreconditionError("Add a payment method first.");
+
+    expect(error.code).toBe(Code.FailedPrecondition);
+    expect(error.rawMessage).toBe("Add a payment method first.");
+    expect(error.findDetails(ErrorInfoSchema)).toEqual([]);
+  });
+
+  it("attaches the reason as an ErrorInfo detail and leaves the message as written", () => {
+    const error = failedPreconditionError("Teams need the Team plan or above.", {
+      reason: "PLAN_UPGRADE_REQUIRED",
+      metadata: { feature: "teams", org_id: "org_1" },
+    });
+
+    expect(error.rawMessage).toBe("Teams need the Team plan or above.");
+    const [info] = error.findDetails(ErrorInfoSchema);
+    expect(info?.reason).toBe("PLAN_UPGRADE_REQUIRED");
+    expect(info?.domain).toBe(ERROR_REASON_DOMAIN);
+    expect(info?.metadata).toEqual({ feature: "teams", org_id: "org_1" });
+  });
+
+  it("delivers the detail to a client across a Connect call", async () => {
+    const transport = createRouterTransport((router) => {
+      router.service(Health, {
+        check: () => {
+          throw failedPreconditionError("A saved payment method is required.", {
+            reason: "PAYMENT_METHOD_REQUIRED",
+            metadata: { org_id: "org_2" },
+          });
+        },
+        list: () => ({ statuses: {} }),
+        watch: async function* () {},
+      });
+    });
+    const client = createClient(Health, transport);
+
+    const caught = await client.check(create(HealthCheckRequestSchema)).then(
+      () => undefined,
+      (error: unknown) => error,
+    );
+
+    expect(caught).toBeInstanceOf(ConnectError);
+    const error = caught as ConnectError;
+    expect(error.code).toBe(Code.FailedPrecondition);
+    expect(error.rawMessage).toBe("A saved payment method is required.");
+    const [info] = error.findDetails(ErrorInfoSchema);
+    expect(info?.reason).toBe("PAYMENT_METHOD_REQUIRED");
+    expect(info?.domain).toBe(ERROR_REASON_DOMAIN);
+    expect(info?.metadata).toEqual({ org_id: "org_2" });
+  });
+});

package/src/pipeline/errors.ts

@@ -9,8 +9,31 @@
  * text can carry storage-engine detail or filesystem paths, and on an
  * anonymous surface that is information disclosure. The cause rides
  * ConnectError.cause for server-side logs; it never crosses the wire.
+ *
+ * A refusal a client should branch on, rather than parse its copy, also
+ * carries a google.rpc.ErrorInfo detail (domain "stigmer.ai"), the
+ * platform's structured-refusal contract that the SDK's getErrorReason
+ * reads. Its reasons are documented on the refusing RPC's proto comment.
+ * The detail rides beside the message and never changes it, so the copy
+ * stays byte-pinned. A detail survives the pipeline and the transport
+ * boundary; only the visitor sanitizer, which rewrites an anonymous
+ * caller's error, drops it (interceptors/error-boundary.ts).
  */
+import { create } from "@bufbuild/protobuf";
 import { Code, ConnectError } from "@connectrpc/connect";
+import { ErrorInfoSchema } from "@stigmer/protos/google/rpc/error_details_pb";
+
+/** The domain every structured refusal reason is issued under. */
+export const ERROR_REASON_DOMAIN = "stigmer.ai";
+
+/**
+ * A machine-readable refusal reason: an UPPER_SNAKE code and the facts a
+ * client needs to act on it, both documented on the refusing RPC.
+ */
+export interface RefusalReason {
+  readonly reason: string;
+  readonly metadata?: Readonly<Record<string, string>>;
+}
 
 /** Go NotFoundError: "%s not found: %s". */
 export function notFoundError(resource: string, id: string): ConnectError {
@@ -32,10 +55,26 @@ export function alreadyExistsError(resource: string, id: string): ConnectError {
 
 /**
  * Go FailedPreconditionError — the system is not in a state required for
- * the operation (vs AlreadyExists, which tells the caller to stop).
+ * the operation (vs AlreadyExists, which tells the caller to stop). With a
+ * reason, the refusal also carries it as an ErrorInfo detail.
  */
-export function failedPreconditionError(message: string): ConnectError {
-  return new ConnectError(message, Code.FailedPrecondition);
+export function failedPreconditionError(
+  message: string,
+  reason?: RefusalReason,
+): ConnectError {
+  if (reason === undefined) {
+    return new ConnectError(message, Code.FailedPrecondition);
+  }
+  return new ConnectError(message, Code.FailedPrecondition, undefined, [
+    {
+      desc: ErrorInfoSchema,
+      value: create(ErrorInfoSchema, {
+        reason: reason.reason,
+        domain: ERROR_REASON_DOMAIN,
+        metadata: { ...reason.metadata },
+      }),
+    },
+  ]);
 }
 
 /** Go AbortedError — retryable conflict (e.g. an in-flight dedupe claim). */