Open Questions Register
The single place open questions and pending decisions live. When work surfaces a question that is not for the current scope — a decision only a human can make, a follow-up, a flagged inconsistency — it lands here as a row, not buried in chat or a PR description.
How it works
- Each question has a stable
OQ-Nid, a scope (product/platform/methodology), who/what raised it, a status, and a resolution note. - Status:
open→answered→ (optionally) graduated to an ADR (technical), a decisions-log entry (product), a GitHub issue (work), or folded into a doc. Resolved rows stay (struck context) so the timeline is visible. - Claude appends here instead of asking inline; the developer scans this list and answers. Answers are recorded in the Resolution column.
Register
| # | Question | Scope | Raised | Status | Resolution / notes |
|---|---|---|---|---|---|
| OQ-1 | ADR-001 says ECS Fargate is the primary compute platform (2024), but the org scan shows EKS is primary today (Terraform EKS + Pulumi + Helm + ArgoCD), ECS secondary (CDK). Write a superseding ADR? | product | taxonomy scan, 2026-06-01 | deferred | Informative — Ariel + Aviad will answer at another time (need the why/when of the EKS move before writing ADR-003). |
| OQ-2 | embedded-widgets (with-calendar + grids in PHP iframes) — its own vertical agent, or a mode of frontend-developer? | platform | agent-taxonomy, 2026-06-01 | answered | Neither — a cross-cutting delivery pattern, not an agent. The widget is React (→ frontend-developer builds the embeddable bundle); the iframe mount is PHP (→ legacy-php-guide); the embed contract (bundle format, postMessage, sizing, token passing) is a shared rule both reference. Removed from the roster as a standalone. Holds as long as the consumer is PHP + iframe. |
| OQ-3 | architect, code-reviewer, process-guardian were partly sourced from agents-developer-kit (now dropped). Do a grounding pass against cursor-rules + the real codebase? | platform | source-hierarchy, 2026-06-01 | answered | Yes — fix existing FIRST, before building new agents. Ariel: don't want to keep coming back to files we created. Grounding pass on the existing agents (+ artifacts) takes priority over Wave A. |
| OQ-4 | Mobile (Flutter) hits the shared backend-services /api/v2 directly — no BFF. Introduce a mobile BFF, or keep the shared API? | product | mobile scan, 2026-06-01 | answered | No BFF. It contradicts Gali's headless-API direction; the version-pinning risk it solves is already covered by /api/v2 versioning + contracts-schema non-breaking discipline. Invest instead in keeping the shared API mobile-friendly (the smoothing fixes already underway). No mobile-bff agent; mobile-developer consumes the shared API. Revisit only if mobile-specific aggregation pain becomes acute. |
| OQ-5 | /continue degrades on unbuilt infra it references by name — lease, state-sync, ai-pr-review.yml, watch-deploy, and the /verify / /pr-review / /done commands. Who builds each, and when? | platform | /continue PR #47, 2026-06-01 | answered | We build all; Claude set the order: (1) /verify + /done + /pr-review (thin wrappers over existing agents — quick, unblock 3 loop steps); (2) state-sync skill; (3) ai-pr-review.yml (needs cloud-bootstrap + Anthropic secret); (4) watch-deploy + incident workflow (needs the obs layer); (5) lease mechanism (heaviest; worktree pattern covers it meanwhile). |
| OQ-6 | Where should ADR-002 (agent taxonomy) live — it is a platform decision, not a product one like ADR-001? | platform | 2026-06-01 | answered | Same docs/adrs/ log + a scope: attribute (product/platform/methodology). Implemented in PR #49. |
| OQ-7 | Open questions keep getting buried in PR/chat prose and are hard to find. | methodology | 2026-06-01 | answered | This register. Claude routes questions here, not inline. |
| OQ-8 | Does Claude need the developer to merge its PRs? | methodology | 2026-06-01 | answered | No — Claude self-merges and may work multiple PRs in parallel (branch protection: 0 approvals, solo mode). |
| OQ-13 | Dedicated incident-response workflow for a post-deploy regression (auto-open ClickUp bug + GitHub issue + Slack on-call, attach signal + suspect commit, propose/trigger rollback) — the "Sentinel" vision. | platform | watch-deploy, 2026-06-04 | deferred | Sequenced after the #99 versioning standard (needs the Layer-4 deploy-record stream + /version contract to consume). Until then watch-deploy flags a regression and routes it to devops-infra/observability for manual handling. Tracked here so it is not lost. |
| OQ-14 | Cross-machine working-state lease (laptop vs a coworker's machine claiming the same task) — needs an external store with TTL/heartbeat (DynamoDB/Redis/API). | platform | session-safety, 2026-06-04 | deferred | Real deferred — possibly never, given the current task-distribution model (one assignee per task; git + ClickUp/GitHub-Issue record handles cross-machine sync). The same-machine case (the actual pain) is solved by the session registry + session-guard + session-drift-check (see session-safety.md). Revisit only if cross-machine claim races actually occur. |