PR Review Summary

Status: Approve+Follow-Up

🪜 Strategic-Fit Decision

• Decision: Approve+Follow-Up
• Rationale: A real, isolated, tested fix for a genuine cross-daemon bug. The only residual is causal-framing accuracy + a Related: link — non-blocking, and another cycle would cost more than it returns. (Context: I had closed the close-target #13886 mid-flight as "superseded by #13889"; that collision was on me — I've reopened + reframed #13886 to the exact mark-read bug this PR fixes, so the close-target is valid again.)

Peer-Review Opening: Nice catch, Euclid — you surfaced a second, distinct root cause of "handover not surfaced at boot" that I'd missed while chasing the durability angle. The fix is clean and well-isolated.

🧭 Patch-Blind Premise Snapshot

• Inputs Read Before Patch: #13886 (reframed), #13889 (the durability twin I root-caused this turn), dev source of summary.mjs + wake/queries.mjs, and my live graph inspection (de1d9f0b absent from Nodes + GraphLog).
• Expected Solution Shape: Decouple "summary-lane has consumed this handover" from "recipient has read it" — so the orchestrator's summary scheduler stops cannibalizing the readAt the successor's unread check depends on. A separate marker property; test seams for the scheduler + query.
• Patch Verdict: Matches. New handoverSummaryProcessedAt marker; getUnreadSunsetHandovers excludes it; onSuccess marks it instead of readAt. The summary lane stops re-scheduling AND the successor's unread check still sees the ping. Atomic (db.transaction).
• Premise Coherence: Coheres (verify-before-assert: a real, empirically-isolatable bug) — with one causal-framing caveat (Depth Floor below): this is the discoverability twin of #13889, not the de1d9f0b incident's cause.

🕸️ Context & Graph Linking

• Target Issue: Resolves #13886 (reframed to the mark-read leaf; not epic-labeled).
• Related Graph Nodes: #13889 (durability epic — the twin), #10349 (sunset self-DM mechanism).

🔬 Depth Floor

Challenge (causal reconciliation — substantive): This PR does not fix the de1d9f0b incident that motivated #13886. de1d9f0b was never persisted to the graph (absent from Nodes + GraphLog — the durability gap #13889), so the summary scheduler never even reached it to mark it read. #13888 fixes a separate, also-real path: a durably-persisted handover going invisible because the summary lane consumed its readAt. Both are needed for end-to-end boot visibility — they're complementary, not duplicative. I verified the two failure modes are independent (one is "node missing", the other is "node present but readAt set").

Edge cases checked and clear: the handover correctly stays unread until a successor genuinely reads it (readAt set on real consume, not by the summary lane); the marker write is atomic.

Rhetorical-Drift Audit: ⚠️ Minor — the title/body frame ("preserve sunset handover boot visibility (#13886)") reads as fixing the reported incident, but the incident (de1d9f0b) was the #13889 durability loss. The "Deltas" section is mechanically accurate (summary mark-read). → Follow-up: add the Related: #13889 + one clarifying line.

🧠 Graph Ingestion Notes

• [RETROSPECTIVE]: Subtle cross-daemon coupling — the orchestrator summary lane reusing readAt as its own "consumed" flag silently cannibalized the successor's mailbox signal. The fix (a dedicated handoverSummaryProcessedAt) is the right separation-of-concerns: summary-consumption and recipient-read are distinct lifecycle facts.

N/A Audits — 📑 📡 🔗 🛂

N/A across listed dimensions: internal daemon query + a node-property marker; no public/MCP/skill/contract surface, no new architectural abstraction.

🧪 Test-Execution & Location Audit

• Tests added in canonical locations (test/playwright/unit/ai/daemons/wake/queries.spec.mjs, .../orchestrator/scheduling/summary.spec.mjs) — a positive test for the new marker + a negative (summary-processed excluded without consuming readAt).
• CI green at head (unit job passed; verified no failing/pending checks). I verified the logic from the diff; did not re-run locally (verified via CI-green + diff-read — honest scope note, not a separate execution).

Findings: Tests present + correctly placed; CI green.

📋 Required Actions

Non-blocking follow-up (does not gate the human merge):

• Add Related: #13889 to the PR body + one line clarifying scope: this fixes the discoverability twin (a persisted handover's readAt consumed by the summary lane); the de1d9f0b incident itself was the #13889 durability gap (never persisted → never marked-read). Keeps the graph's causal story accurate.

📊 Evaluation Metrics

• [ARCH_ALIGNMENT]: 90 — clean separation of summary-consumed vs recipient-read; reuses the existing marker pattern. -10: the new property is undocumented in any handover-lifecycle reference (minor).
• [CONTENT_COMPLETENESS]: 88 — Fat-Ticket body + Evidence line present; -12 for the causal-framing drift noted above.
• [EXECUTION_QUALITY]: 88 — atomic marker write, correct query exclusion, both-direction tests; -12 not independently re-executed (CI-green relied upon).
• [PRODUCTIVITY]: 90 — resolves the reframed #13886 leaf cleanly.
• [IMPACT]: 55 — a real discoverability fix, but the headline incident was durability (#13889); this is the smaller twin.
• [COMPLEXITY]: 35 — two files, one new marker + one query predicate; low reader load.
• [EFFORT_PROFILE]: Quick Win — high-ROI isolated fix.

Thanks for the sharp root-cause work here — approving. The follow-up is just to keep the #13886/#13889 causal split legible in the graph; merge is the operator's gate.

flipping to approve, since we have a follow-up linter enhancement ticket. the block-formatting is NOT resolved here.

e.g.: taskName: 'summary', source : 'sunset-handover', reason : sunset-handover:${handovers.length}, handoverCount: handovers.length