Adopt Claude Code Remote Control + mobile push notifications

[tractorjuice]

Summary

Claude Code v2.1.110+ supports Remote Control (drive a local session from claude.ai/code or the mobile app) and mobile push notifications (Claude pushes when long-running work finishes or needs a decision). ArcKit's workload — long-running research agents, stale-artifact monitoring, multi-stage approval cycles — is a strong fit. This issue tracks adopting the pattern in guides and, where helpful, in plugin metadata.

Min Claude Code floor today is v2.1.117 (per project_min_claude_code_version.md), so the v2.1.110 requirement is already covered.

Use cases (ordered by leverage)

1. Long agent runs

Heaviest workloads — /arckit.research, /arckit.datascout, /arckit.aws-research, /arckit.azure-research, /arckit.gcp-research, /arckit.grants, and the overnight autoresearch loop. All do dozens of WebSearch/WebFetch/MCP calls and frequently exceed 10 min.

• Recommend claude remote-control (server mode) for these.
• Tell users to enable /config → Push when Claude decides so the phone pings on completion or when the agent hits a decision point (build vs buy direction, vendor shortlist confirmation).
• Replaces tail-the-log workflows; complements the existing Monitor tool guidance.

2. Stale-artifact monitor

arckit-claude/.claude-plugin/plugin.json already declares the stale-artifact-scan monitor (runs detect-stale-artifacts.sh at session start, emits one line per overdue artifact). Pairing it with RC + push means owners get a phone nudge when reviews lapse without sitting at the terminal.

3. Document Control approval cycles

Status transitions (DRAFT → IN_REVIEW → APPROVED) and Next Review Date triggers map naturally to "notify me when this needs sign-off." No code change required — a usage pattern in approval/review guides telling owners to launch the session via RC and ask for a push.

Proposed work

• Add a "Remote Control + push" section to docs/guides/autoresearch.md (overnight scoring loops are the canonical fit)
• Add the same recommendation to the research-heavy command guides (research.md, datascout.md, aws-research.md, azure-research.md, gcp-research.md, grants.md)
• Document RC + push pairing for the stale-artifact-scan monitor in the plugin README and any review-cycle guide
• Add a short note to CLAUDE.md under "Plugin Monitors" / "Monitor Tool" pointing at the RC pattern
• Decide whether to bump the documented min Claude Code floor message to mention RC explicitly (currently v2.1.117, already covers v2.1.110 RC requirement)

No converter / extension changes — RC is Claude Code only and doesn't need to propagate to Codex/Gemini/OpenCode/Copilot.

Tradeoffs / constraints

• Pro/Max plans only; Team/Enterprise needs the admin toggle in claude.ai admin settings; API keys unsupported.
• Local claude process must stay running. ~10 min network outage ends the session.
• Push has a single on/off — no per-event routing. A chatty agent could over-notify, so guides should call this out.
• /ultraplan disconnects RC, so RC does not compose with /ultrareview-style flows.
• Bedrock / Vertex / Foundry providers unsupported.

Out of scope

• Building any new commands or hooks tied to RC. The feature is mostly user-config + documentation; the leverage comes from telling users where it fits, not from new ArcKit code.