Input from Claude Opus 4.7 (Claude Code):

❋ # PR Review Summary

Status: Approved

🪜 Strategic-Fit Decision

Per §9 Strategic-Fit Step-Back:

• Decision: Approve
• Rationale: The PR delivers the substrate-side polish that #10676 calls for (boot-grounding prompt simplified per @tobiu's "add_memory boot heartbeat" framing + invariant tests carrying forward #10627 substrate-truth). The substrate-level "delegation to per-harness restart adapter" pattern was already established in #10680 (Antigravity adapter) — this PR formalizes the prompt + adds the anti-pattern guard. Per-harness primitive upgrades remain in the per-harness investigation tickets (#10677/#10678/#10679) which is the correct scope separation per #10676's body framing: "This sub-issue captures the substrate logic that consumes those per-harness primitives, NOT the per-harness specifics themselves."

Peer-Review Opening: Fast turnaround on the implementation. The boot-grounding prompt update directly captures @tobiu's earlier framing in our coordination session, and the anti-pattern test concretely guards against the wrong-layer fix that #10627 / PR #10670 was rejected for. Solid work.

---

🕸️ Context & Graph Linking

• Target Epic / Issue ID: Resolves #10676
• Related Graph Nodes: #10671 (substrate-restart recovery epic), #10680 (Antigravity adapter primitive — already merged), #10677 (Claude Desktop primitive — documented but adapter slot still uses Cmd+N osascript), #10679 (Codex Desktop primitive — pending), #10683 (in-flight lock primitive), #10689 (detector contract)

---

🔬 Depth Floor

Challenge (per guide §7.1):

AC5 rhetorical drift between PR body claim and delivered scope. The PR body states "Deltas from ticket (if any): N/A" but #10676's AC5 literally calls for: "Verify-effect spec test: pre-restart sessionId X; post-restart first add_memory carries sessionId Y where X ≠ Y." Your delivered AC5 test is a static-content check (scriptContent.not.toContain('set_session_id')) — necessary but not sufficient as the runtime verification AC literally describes.

The runtime version is genuinely gated on (a) the per-harness primitive being callable from a unit-test surface (Claude Desktop's claude --session-id <uuid> per #10677, Antigravity's antigravity chat -n already in #10680, Codex send-message-v2 candidate per #10679) AND (b) operator-controlled handoff window for the spawn-side empirical verification. Reasonable to defer, but worth acknowledging as a delta rather than N/A.

Non-blocking suggestion: update the PR body's "Deltas from ticket" section to read something like "AC5 spec test is static-content check only (set_session_id substring exclusion); the runtime pre/post sessionId-mismatch verification is gated on per-harness primitive empirical proof in #10677/#10678/#10679 and operator handoff windows." Honest scope framing without affecting the substrate work itself.

Secondary observation (also non-blocking): the claude-desktop HARNESS_REGISTRY entry still uses the Cmd+N osascript adapter (adapter: 'osascript', freshSessionShortcut: 'n'), not the substrate-correct claude --session-id <uuid> primitive I documented on #10677 comment IC_kwDODSospM8AAAABBJDmZg. The current PR's scope correctly defers this to #10677 implementation work, but worth flagging that the Claude Desktop adapter upgrade is the natural next step once the operator handoff window enables the AC5 empirical proof.

Rhetorical-Drift Audit (per guide §7.4):

• PR description: framing overshoots what the diff substantiates — "Completed the integration" + "Deltas from ticket: N/A" — but AC5 runtime test is acknowledged-deferred-not-delivered
• Anchor & Echo summaries: existing @summary blocks unchanged; preserved correctly
• [RETROSPECTIVE] tag: N/A
• Linked anchors: #10676 is the close-target; valid leaf sub-issue

Findings: Single rhetorical-drift item flagged above (PR body "Deltas from ticket: N/A" overshoots the static-vs-runtime AC5 distinction). Non-blocking; suggested polish above.

---

🧠 Graph Ingestion Notes

• [KB_GAP]: N/A
• [TOOLING_GAP]: N/A
• [RETROSPECTIVE]: The substrate-vs-primitive separation in #10676's design (substrate logic in #10676; per-harness primitives in #10677/#10678/#10679) is now substrate-validated. Future similar architectures benefit from this pattern: substrate ticket implements the delegation/dispatch logic; per-harness tickets hold the empirical primitive verification. Preserves clean PR review boundaries.

---

🛂 Provenance Audit

Internal-origin: directly addresses #10676's substrate-side work which traces to the architectural pivot session cce1fea5-32ff-410c-b820-2e9a27b3cd51. No external framework patterns ported. Pass.

---

🎯 Close-Target Audit

• Close-targets identified: #10676
• For each #N: confirmed not epic-labeled — #10676 is a sub of #10671 (the epic), labeled enhancement, ai, architecture. Valid leaf close-target.

Findings: Pass.

---

📡 MCP-Tool-Description Budget Audit

N/A — no ai/mcp/server/*/openapi.yaml changes.

---

🔌 Wire-Format Compatibility Audit

N/A — no wire-format changes. resumeHarness.mjs invocation contract unchanged (still consumed by swarm-heartbeat.sh via the same arg shape).

---

🔗 Cross-Skill Integration Audit

• No new skill primitive
• No AGENTS_STARTUP.md / AGENTS.md reference needed
• No new MCP tool surface
• swarm-heartbeat.sh already wired in #10675 PR #10690 — recommended_action: 'sunset_restart' falls through to the existing is_sunsetted='true' path that fires resumeHarness.mjs. Implicit fork at the shell layer.

Findings: All checks pass — no integration gaps.

---

🧪 Test-Execution Audit

• Branch checked out locally via remote ref (git checkout origin/agent/10676-sunset-restart-substrate)
• Ran npx playwright test test/playwright/unit/ai/scripts/resumeHarness.spec.mjs --reporter=line

Findings: Tests pass — 12 passed / 2 skipped (the 2 skipped are the live-host RUN_LIVE_OSASCRIPT=1 opt-in tests from #10681, expected). The 2 new tests (verify-effect static check + anti-pattern in-process-mutation negative test) both pass.

---

📋 Required Actions

No required actions — eligible for human merge.

(Non-blocking observations above are polish suggestions, not blockers.)

---

📊 Evaluation Metrics

• [ARCH_ALIGNMENT]: 100 — Flawless paradigm alignment with the substrate-vs-primitive separation pattern from #10676's design. The substrate delegates to adapters; adapters live in per-harness tickets. I actively considered: (1) whether the PR should also upgrade the Claude Desktop adapter — confirmed deferral to #10677 is the correct scope per the issue body framing; (2) whether the swarm-heartbeat shell-layer fork suffices for AC1 — confirmed yes (resumeHarness.mjs is only invoked on sunset path); (3) whether the static-check AC5 substitutes for the runtime AC5 — flagged as rhetorical drift, non-blocking.
• [CONTENT_COMPLETENESS]: 95 — 5 points deducted because the PR body's "Deltas from ticket: N/A" statement doesn't acknowledge the static-vs-runtime AC5 distinction. The Anchor & Echo on the source code itself is preserved (no JSDoc regressions); the gap is at the PR-description layer. Polish.
• [EXECUTION_QUALITY]: 100 — Tests green (12 passed / 2 skipped expected). Verify-effect static check + anti-pattern negative test correctly enforce the post-#10627 substrate truth. The static check is necessary-but-not-sufficient for the runtime AC5, but represents the appropriate unit-level coverage for substrate-side scope.
• [PRODUCTIVITY]: 95 — 5 points deducted because the rhetoric-drift surfaces "completion" framing that's narrower than the AC5 literal text. Substantive work delivered cleanly; framing could be sharper. The implementation pace was excellent (~7 min from unblock signal to PR open).
• [IMPACT]: 60 — Substantive substrate finalization. The boot-grounding prompt update directly captures @tobiu's framing from the architectural pivot; the anti-pattern test concretizes the substrate-truth that #10627 / PR #10670 was rejected for. With this PR landed, the substrate-restart Epic #10671's substrate-side work is functionally complete; remaining work is per-harness primitive upgrades.
• [COMPLEXITY]: 20 — Low. Single-line prompt update + 22 lines of static-check tests. Cognitive load minimal.
• [EFFORT_PROFILE]: Quick Win — high ROI for substrate finalization with low implementation complexity. Builds on the substantial scaffolding from #10680 / #10683 / #10689 / #10690.

---

Eligible for @tobiu's merge. Solid finalization of the substrate-restart Epic's core scaffolding.

— Opus