@@ -0,0 +1,16 @@
+---
+'@objectstack/service-automation': patch
+---
+
+fix(service-automation): a `connector_action` dispatch to a held connector instance names what recovers it, instead of saying the platform retries it
+
+Clause-②: no
+Metadata-protocol: none
+
+A degraded connector instance that the platform does not retry (it is **held**) used to answer a dispatch with "Dispatch is unavailable until that is resolved; the platform retries automatically." Nothing retried it. Now the refusal names the fix:
+
+- **A configuration fault met on a retry** (#27992), or **a credential written into the declaration that the upstream refused**: "Dispatch is unavailable until its declaration is edited (or the server restarts); the platform does not retry it."
+- **A stored credential the upstream refused** (#27237): "Dispatch is unavailable until credential '<ref>' is set again (or the server restarts); the platform does not retry it."
+- **An instance on the backoff** (an unreachable upstream, an environment or vault credential the upstream refused) keeps "the platform retries automatically".
+
+The engine's degraded-connector record carries the hold: `registerDegradedConnector` takes an optional fifth argument, `RegisteredConnector` gains an optional `degradedHold`, and `AutomationEngine.getConnectorDegradedHold(name)` reads it. The engine's `Connector registered DEGRADED` warning no longer says the materializer retries a held instance. A degraded reason that ends with a full stop no longer gets a second one in the refusal.
@@ -454,6 +454,7 @@ retry):
 | Instance `name` collides with a plugin-registered connector | **Hard boot error** — there is no silent precedence between the two worlds |
 | stdio transport without host opt-in | **Hard boot error** with the `declarativeStdio` fix inline — a security rejection is never retried |
 | MCP server unreachable / `tools/list` fails; remote spec URL unreachable or transiently failing (408/429/5xx) | **Not fatal**: the instance registers **degraded** (visible on `GET /api/v1/automation/connectors` with the reason), dispatches fail with that reason, and the platform retries on a 5s→5min backoff — recovery is automatic |
+| A retry of a degraded instance reaches the upstream and meets a configuration fault (an `include` naming a tool or operation the upstream does not offer) | **Not fatal**: the instance stays **degraded**, now with the configuration reason, and is **not retried** — dispatches fail saying to edit its declaration; an edit of the entry (or a restart) re-attempts it |
 | Wrong spec URL (other 4xx) or unparseable document | Configuration fault — **hard boot error** |
 | A bad entry arrives via **runtime reload** (Studio publish / dev reload) | Logged and **skipped**, never crashing the live server; a changed instance's old connector keeps serving until its replacement materializes |
 
 
@@ -4,6 +4,7 @@ import type { PluginContext } from '@objectstack/core';
 import { defineActionDescriptor } from '@objectstack/spec/automation';
 import type { AutomationEngine, ConnectorActionContext } from '../engine.js';
 import { refuseNode } from '../guard-refusal.js';
+import { describeHold } from '../plugin-extension-api.js';
 import { resolveValueSlotMap } from './value-slot-map.js';
 
 /**
@@ -63,17 +64,24 @@ export function registerConnectorNodes(engine: AutomationEngine, ctx: PluginCont
             const handler = engine.resolveConnectorAction(cfg.connectorId, cfg.actionId);
             if (!handler) {
                 // A degraded declarative instance (#3017) is registered but has no
-                // handlers — say WHY dispatch fails and that recovery is automatic,
-                // instead of the generic wiring hint below. The reason names the
-                // cause: an unreachable upstream, or (#27237) a credential that
-                // is not set.
+                // handlers — say WHY dispatch fails and what recovers it, instead
+                // of the generic wiring hint below. The reason names the cause: an
+                // unreachable upstream, or (#27237) a credential that is not set or
+                // was refused. [#28016] Recovery is automatic only on the backoff:
+                // a held instance (a configuration fault met on a retry, a refused
+                // credential the platform will not re-send) names its fix instead.
                 const degraded = engine.getConnectorDegradedReason(cfg.connectorId);
                 if (degraded) {
+                    const hold = engine.getConnectorDegradedHold(cfg.connectorId);
+                    // A reason that is a full sentence keeps one full stop.
+                    const reason = degraded.replace(/\.\s*$/, '');
                     return {
                         success: false,
                         error:
-                            `connector_action '${node.id}': connector '${cfg.connectorId}' is degraded — ${degraded}. `
-                            + `Dispatch is unavailable until that is resolved; the platform retries automatically.`,
+                            `connector_action '${node.id}': connector '${cfg.connectorId}' is degraded — ${reason}. `
+                            + (hold
+                                ? `Dispatch is unavailable until ${describeHold(hold)} (or the server restarts); the platform does not retry it.`
+                                : `Dispatch is unavailable until that is resolved; the platform retries automatically.`),
                     };
                 }
                 return {
 
@@ -722,3 +722,64 @@ describe('[#27237] contract review: an encoded spelling of the value is withheld
         }
     });
 });
+
+/** [#28016] Run a one-node flow dispatching to `billing.request` and return its refusal text. */
+async function dispatchRefusal(engine: AutomationEngine): Promise<string> {
+    engine.registerFlow('call_billing', {
+        name: 'call_billing',
+        label: 'Call billing',
+        type: 'autolaunched',
+        nodes: [
+            { id: 'start', type: 'start', label: 'Start' },
+            { id: 'call', type: 'connector_action', label: 'Request', connectorConfig: { connectorId: 'billing', actionId: 'request', input: {} } },
+            { id: 'end', type: 'end', label: 'End' },
+        ],
+        edges: [
+            { id: 'e1', source: 'start', target: 'call' },
+            { id: 'e2', source: 'call', target: 'end' },
+        ],
+    } as never);
+    const result = await engine.execute('call_billing');
+    expect(result.success).toBe(false);
+    return String(result.error);
+}
+
+describe('[#28016] a dispatch to an instance held on a refused stored credential says to set it again', () => {
+    it('refused at materialization (the mcp shape): set the credential again; nothing retries it', async () => {
+        const { engine, provider } = await boot([billing()]);
+        provider.state.rejectAtMaterialization = true;
+        await engine.setConnectorCredential({ ref: REF, value: DUMMY_VALUE, context: ADMIN });
+
+        const error = await dispatchRefusal(engine);
+        expect(error).toContain("connector 'billing' is degraded — ");
+        expect(error).toContain(`until credential '${REF}' is set again (or the server restarts); the platform does not retry it.`);
+        expect(error).not.toMatch(/retries automatically/);
+        expect(error).not.toContain('.. Dispatch');
+        expect(error.includes(DUMMY_VALUE)).toBe(false);
+    });
+
+    it('refused on a call (the rest / openapi shape): set the credential again; nothing retries it', async () => {
+        const { engine, provider, call } = await boot([billing()]);
+        await engine.setConnectorCredential({ ref: REF, value: DUMMY_VALUE, context: ADMIN });
+        provider.state.callStatus = 401;
+        await call();
+
+        const error = await dispatchRefusal(engine);
+        expect(error).toContain(`until credential '${REF}' is set again (or the server restarts); the platform does not retry it.`);
+        expect(error).not.toMatch(/retries automatically/);
+        expect(error).not.toContain('.. Dispatch');
+    });
+
+    it('control: an environment value refused on a call is on the backoff, and the refusal says the platform retries it', async () => {
+        process.env[REF] = DUMMY_ENV_VALUE;
+        envKeys.push(REF);
+        const { engine, provider, call } = await boot([billing()]);
+        provider.state.callStatus = 403;
+        await call();
+
+        const error = await dispatchRefusal(engine);
+        expect(error).toContain('The instance is retried with backoff.');
+        expect(error).toContain('Dispatch is unavailable until that is resolved; the platform retries automatically.');
+        expect(error).not.toMatch(/does not retry/);
+    });
+});
@@ -0,0 +1,164 @@
+// Copyright (c) 2026 ObjectStack. Licensed under the Apache-2.0 license.
+//
+// #28016 — a `connector_action` dispatch to a degraded instance says what
+// recovers it. An instance on the #3017 backoff recovers by itself, so its
+// refusal says the platform retries it. A HELD instance is not retried:
+// a configuration fault met on a backoff retry (#27992), or a credential the
+// upstream refused when only an edit of the declaration changes what it sends
+// (#27237). Its refusal names the fix and never says it retries. The stored-
+// credential hold (set the credential again) is pinned beside its door, in
+// connector-credential-door.test.ts.
+
+import { describe, it, expect, afterEach, vi } from 'vitest';
+import { LiteKernel, ObjectLogger } from '@objectstack/core';
+import type { Connector, ConnectorProviderFactory } from '@objectstack/spec/integration';
+import {
+    ConnectorCredentialRejectedError,
+    ConnectorUpstreamUnavailableError,
+} from '@objectstack/spec/integration';
+import { AutomationServicePlugin, DECLARATIVE_RETRY_BASE_MS } from './plugin.js';
+import { AutomationEngine } from './engine.js';
+
+type Upstream = 'down' | 'refuses-include' | 'refuses-credential';
+
+/** An mcp-shaped provider: the marker errors for down / refused credential, a plain Error for a bad `include`. */
+function makeProvider(upstream: Upstream) {
+    const state = { upstream, calls: 0 };
+    const factory: ConnectorProviderFactory = (ctx) => {
+        state.calls++;
+        if (state.upstream === 'down') {
+            throw new ConnectorUpstreamUnavailableError(
+                `connector-mcp provider: connector '${ctx.name}' could not reach its MCP server: connect ECONNREFUSED`,
+            );
+        }
+        if (state.upstream === 'refuses-credential') {
+            throw new ConnectorCredentialRejectedError(
+                `connector-mcp provider: connector '${ctx.name}' was refused by its upstream (HTTP 401).`,
+                401,
+            );
+        }
+        throw new Error(
+            `connector-mcp provider: connector '${ctx.name}' providerConfig.include names 'nope', which the MCP server does not list.`,
+        );
+    };
+    return { factory, state };
+}
+
+const ENTRY = { name: 'upstream', label: 'upstream', type: 'api', provider: 'mcp', providerConfig: { url: 'https://upstream.example', include: ['nope'] } };
+
+async function boot(factory: ConnectorProviderFactory) {
+    const kernel = new LiteKernel({ logger: { level: 'silent' } } as never);
+    kernel.use(new AutomationServicePlugin());
+    kernel.use({
+        name: 'test.harness',
+        type: 'standard' as const,
+        version: '1.0.0',
+        dependencies: ['com.objectstack.service-automation'],
+        async init(ctx: any) {
+            ctx.registerService('objectql', {
+                registry: { listItems: (type: string) => (type === 'connector' ? [ENTRY] : []) },
+            });
+            ctx.getService('automation').registerConnectorProvider('mcp', factory);
+        },
+        async start() {},
+    } as never);
+    await kernel.bootstrap();
+    return { kernel, engine: kernel.getService('automation') as AutomationEngine };
+}
+
+/** Run a one-node flow dispatching to `upstream.ping` and return its refusal text. */
+async function dispatchRefusal(engine: AutomationEngine): Promise<string> {
+    engine.registerFlow('call_upstream', {
+        name: 'call_upstream',
+        label: 'Call upstream',
+        type: 'autolaunched',
+        nodes: [
+            { id: 'start', type: 'start', label: 'Start' },
+            { id: 'call', type: 'connector_action', label: 'Ping', connectorConfig: { connectorId: 'upstream', actionId: 'ping', input: {} } },
+            { id: 'end', type: 'end', label: 'End' },
+        ],
+        edges: [
+            { id: 'e1', source: 'start', target: 'call' },
+            { id: 'e2', source: 'call', target: 'end' },
+        ],
+    } as never);
+    const result = await engine.execute('call_upstream');
+    expect(result.success).toBe(false);
+    return String(result.error);
+}
+
+describe('#28016 — a dispatch to a held connector instance names its fix, not a retry', () => {
+    afterEach(() => {
+        vi.useRealTimers();
+    });
+
+    it('a configuration fault met on the backoff retry: edit the declaration; nothing retries it', async () => {
+        vi.useFakeTimers();
+        const { factory, state } = makeProvider('down');
+        const { kernel, engine } = await boot(factory);
+        state.upstream = 'refuses-include';
+        await vi.advanceTimersByTimeAsync(DECLARATIVE_RETRY_BASE_MS);
+        expect(state.calls).toBe(2);
+
+        const error = await dispatchRefusal(engine);
+        expect(error).toContain("connector 'upstream' is degraded — ");
+        expect(error).toContain("include names 'nope'");
+        expect(error).toContain('until its declaration is edited (or the server restarts); the platform does not retry it.');
+        expect(error).not.toMatch(/retries automatically/);
+        expect(error).not.toContain('.. Dispatch');
+
+        await kernel.shutdown();
+    });
+
+    it('a credential written into the declaration, refused by the upstream: edit the declaration; nothing retries it', async () => {
+        const { factory } = makeProvider('refuses-credential');
+        const { kernel, engine } = await boot(factory);
+        expect(engine.getConnectorDegradedReason('upstream')).toMatch(/refused by its upstream \(HTTP 401\)/);
+
+        const error = await dispatchRefusal(engine);
+        expect(error).toContain('until its declaration is edited (or the server restarts); the platform does not retry it.');
+        expect(error).not.toMatch(/retries automatically/);
+        expect(error).not.toContain('.. Dispatch');
+
+        await kernel.shutdown();
+    });
+
+    it('control: an unreachable upstream is on the backoff, and the refusal says the platform retries it', async () => {
+        vi.useFakeTimers();
+        const { factory, state } = makeProvider('down');
+        const { kernel, engine } = await boot(factory);
+        await vi.advanceTimersByTimeAsync(DECLARATIVE_RETRY_BASE_MS);
+        expect(state.calls).toBe(2);
+
+        const error = await dispatchRefusal(engine);
+        expect(error).toContain('could not reach its MCP server');
+        expect(error).toContain('Dispatch is unavailable until that is resolved; the platform retries automatically.');
+        expect(error).not.toMatch(/does not retry/);
+
+        await kernel.shutdown();
+    });
+});
+
+describe('#28016 — the degraded record carries the hold', () => {
+    const husk = { name: 'billing', label: 'billing', type: 'api', authentication: { type: 'none' }, actions: [] } as unknown as Connector;
+
+    it('a held degrade stores what re-attempts it and its warn says it is not retried; an unheld one clears it', () => {
+        const logger = new ObjectLogger({ level: 'silent' });
+        const warn = vi.spyOn(logger, 'warn');
+        const engine = new AutomationEngine(logger);
+        const lastWarn = () => String(warn.mock.calls[warn.mock.calls.length - 1]?.[0]);
+
+        engine.registerDegradedConnector(husk, 'refused', 'declarative', undefined, { kind: 'credential', ref: 'BILLING_TOKEN' });
+        expect(engine.getConnectorDegradedHold('billing')).toEqual({ kind: 'credential', ref: 'BILLING_TOKEN' });
+        expect(lastWarn()).toContain(
+            "the materializer does not retry it until credential 'BILLING_TOKEN' is set again (or a restart).",
+        );
+        expect(lastWarn()).not.toContain('retries with backoff');
+
+        // The next degrade of the same name replaces the record: on the backoff, no hold.
+        engine.registerDegradedConnector(husk, 'could not reach it', 'declarative');
+        expect(engine.getConnectorDegradedHold('billing')).toBeUndefined();
+        expect(lastWarn()).toContain('the materializer retries with backoff.');
+        expect(engine.getConnectorDegradedHold('unregistered')).toBeUndefined();
+    });
+});
@@ -667,8 +667,27 @@ export interface RegisteredConnector {
     readonly state: ConnectorState;
     /** Why the connector is degraded — set only when `state` is `degraded`. */
     readonly degradedReason?: string;
+    /**
+     * [#28016] Set only on a degraded connector the platform does NOT retry:
+     * what re-attempts it instead. Absent on an instance on the #3017 backoff.
+     */
+    readonly degradedHold?: ConnectorHold;
 }
 
+/**
+ * [#28016] What re-attempts a **held** degraded instance — one neither the
+ * retry timer nor an unchanged reload re-attempts (a restart always does):
+ *
+ *  - `declaration`: an edit of its declaration — a configuration fault met on
+ *    a retry (#27992), or a credential written into the declaration that the
+ *    upstream refused (#27237);
+ *  - `credential`: setting the stored credential `ref` again — a value stored
+ *    through the credential door that the upstream refused (#27237).
+ */
+export type ConnectorHold =
+    | { readonly kind: 'declaration' }
+    | { readonly kind: 'credential'; readonly ref: string };
+
 // The connector **provider** contract (ADR-0097) — ConnectorProviderFactory,
 // ConnectorProviderContext, ConnectorMaterialization — lives in
 // `@objectstack/spec/integration` so a connector plugin can implement a factory
@@ -3982,20 +4001,24 @@ export class AutomationEngine implements IAutomationService {
      * @param cause the thrown value behind the degrade, for the log record's
      *   structured `meta`. Optional — a caller with no thrown value in hand
      *   still gets `degradedReason` in the record.
+     * @param hold [#28016] set when the platform will NOT retry the instance:
+     *   what re-attempts it instead, stored as `degradedHold` and named by a
+     *   `connector_action` refusal. Omitted, the instance is on the backoff.
      */
     registerDegradedConnector(
         def: Connector,
         reason: string,
         origin: ConnectorOrigin = 'declarative',
         cause?: unknown,
+        hold?: ConnectorHold,
     ): void {
         // The body lives in ./plugin-extension-api.ts (#25222).
         const host: ThisParameterType<typeof registerDegradedConnector> = {
             assertSameOriginOrFree: (...a) => this.assertSameOriginOrFree(...a),
             connectors: this.connectors,
             logger: this.logger,
         };
-        registerDegradedConnector.call(host, def, reason, origin, cause);
+        registerDegradedConnector.call(host, def, reason, origin, cause, hold);
     }
 
     /** Enforce the ADR-0097 §4 two-sources-of-truth rule; warn on same-origin replace. */
@@ -4142,6 +4165,17 @@ export class AutomationEngine implements IAutomationService {
         return reg?.state === 'degraded' ? (reg.degradedReason ?? 'upstream unavailable') : undefined;
     }
 
+    /**
+     * [#28016] What re-attempts a held degraded connector, or `undefined` when
+     * it is not held — `ready`, unregistered, or degraded on the #3017 backoff.
+     * Lets the `connector_action` node say "retries automatically" only when
+     * that is true.
+     */
+    getConnectorDegradedHold(name: string): ConnectorHold | undefined {
+        const reg = this.connectors.get(name);
+        return reg?.state === 'degraded' ? reg.degradedHold : undefined;
+    }
+
     /**
      * Register a connector **provider factory** (ADR-0097 §2). A connector
      * plugin (e.g. `@objectstack/connector-openapi`) calls this at `init()` under