I4: Engage upstream — comment on TerriaJS#7742 + author sonnet-reviewer subagent

[HichemMokni]

Phase 1 — Engagement

Two parallel sub-tasks:

A. Comment on TerriaJS#7742. Tag @ShubhamSharmaFAO and @zoran995. State NextAV's use case (Sentinel-2 SR over GeoServer GWC; Qatar Energy demo Q2 2026). Frame as co-authorship offer, not takeover. Include a 5-business-day window: "If you'd prefer we open the PR with credit to you, we can do that — please reply by [DATE+5BD]."

B. Author the sonnet-reviewer subagent. Create /home/nextav/Agents/Engineering/.claude/agents/sonnet-reviewer.md. Sonnet-backed second-opinion reviewer; design + edge-cases focus; non-overlapping with guardian (Opus, security/quality). Used for two test-drive PRs (I5 + I6) before high-stakes ones (I7, I13).

Also answer @zoran995's open design questions in the issue thread before any code lands:

• (a) Yes, follow WMS pattern of imagery-provider-per-time.
• (b) Yes, reuse initialTimeSource via DiscretelyTimeVaryingMixin.

Plan reference: Phase 1 step 2; Roles & Review > Sonnet-reviewer subagent
Blocks: I5–I8 (subagent must exist before agent-only review of impl PRs).

[HichemMokni]

I4 complete.

Part A: engagement comment posted on TerriaJS#7742 (TerriaJS#7742 (comment)) tagging @ShubhamSharmaFAO and @zoran995, framing as co-authorship offer. Posted alongside opening of upstream PR TerriaJS#7835.

Part B: sonnet-reviewer subagent shipped in nextavllc/engineering-agents commit 3cc3548 (.claude/agents/sonnet-reviewer.md). Used in production for both PR #11 round 1+2 and PR #12 round 1+2 reviews. The disagreement protocol (3-strike escalation to Hichem) was tested implicitly when the implementer pushed back on a wrong reviewer assumption about createDiscreteTimes bucket selection — reviewer accepted the correction in round 2.