Summary

• Fixes [Bug]: 当tts设置为always时，cron会投递失败，但是结果会显示已投递 #92816, where a QQBot cron/TTS media send could be counted as delivered even when the adapter returned no usable delivery identity.
• The shared outbound media path now suppresses media adapter results that do not include a usable delivery id.
• This does not change QQBot sending itself; it only changes how the shared delivery layer accounts for an empty media result.

Linked context

Closes #92816

Real behavior proof (required for external PRs)

• Behavior or issue addressed: QQBot-shaped media results with messageId: "" are no longer counted as delivered.

• Real environment tested: Local OpenClaw checkout on this PR branch, Node 24.14.0, using the real outbound delivery code path with a QQBot-shaped adapter response.

• Exact steps or command run after this patch:

  pnpm tsx --eval 'import { deliverOutboundPayloads } from "./src/infra/outbound/deliver.ts";
  import { setActivePluginRegistry } from "./src/plugins/runtime.ts";
  import { createOutboundTestPlugin, createTestRegistry } from "./src/test-utils/channel-plugins.ts";
  
  void (async () => {
    setActivePluginRegistry(
      createTestRegistry([
        {
          pluginId: "qqbot",
          source: "manual-proof",
          plugin: createOutboundTestPlugin({
            id: "qqbot",
            outbound: {
              deliveryMode: "direct",
              sendText: async () => ({ channel: "qqbot" as const, messageId: "text-1" }),
              sendMedia: async () => ({ channel: "qqbot" as const, messageId: "" }),
            },
          }),
        },
      ]),
    );
    const outcomes: unknown[] = [];
    const results = await deliverOutboundPayloads({
      cfg: {},
      channel: "qqbot",
      to: "qq-target",
      payloads: [{ text: "voice caption", mediaUrl: "file:///tmp/cron-tts.mp3" }],
      skipQueue: true,
      onPayloadDeliveryOutcome: (outcome) => outcomes.push(outcome),
    });
    console.log(JSON.stringify({ results, outcomes }, null, 2));
  })();'

• Evidence after fix:

  {
    "results": [],
    "outcomes": [
      {
        "index": 0,
        "status": "suppressed",
        "reason": "adapter_returned_no_identity"
      }
    ]
  }

• Observed result after fix: The outbound delivery result stays empty when the media adapter returns no delivery identity, so cron/TTS delivery accounting no longer sees that send as delivered.

• What was not tested: I did not run a live QQBot cron/TTS delivery. I do not have QQBot credentials available in this checkout.

• Proof limitations or environment constraints: The live service was not exercised; the proof covers the shared delivery boundary changed by this PR.

• Before evidence (optional but encouraged): The new regression failed before the fix because the media branch counted the empty adapter response as delivered.

Tests and validation

• pnpm test src/infra/outbound/deliver.test.ts -- --reporter=verbose
• pnpm test extensions/qqbot/src/channel.message-adapter.test.ts -- --reporter=verbose
• pnpm test src/cron/isolated-agent/delivery-dispatch.double-announce.test.ts -- --reporter=verbose -t "applies TTS directives before direct cron announce delivery and mirrors spoken text|keeps unresolved message-tool delivery out of delivered status"
• pnpm format:check -- src/infra/outbound/deliver.ts src/infra/outbound/deliver.test.ts
• git diff --check
• pnpm tsgo:core:test
• Local autoreview completed with no accepted/actionable findings.

Regression coverage added: src/infra/outbound/deliver.test.ts now covers a media adapter result with no delivery identity.

Risk checklist

Did user-visible behavior change? Yes

Did config, environment, or migration behavior change? No

Did security, auth, secrets, network, or tool execution behavior change? No

Highest-risk area: shared outbound media delivery accounting.

How is that risk mitigated? The change is limited to delivery result accounting. Media sends with a real delivery identity are still counted; responses without an identity are treated as not delivered.

Current review state

Next action: maintainer review and CI.

Waiting on: reviewer feedback and CI.

Bot/reviewer comments addressed: none yet.

AI-assisted (Claude). The proof above was run manually.