Enhancement: surface local resource pressure (RAM/CPU/disk) below the input bar

Summary

When Claude Code agents (subagents, parallel tool calls, MCP servers, long-running shells) start to bloat memory or saturate CPU/disk on the host machine, there's no in-app signal. The session keeps feeling fast right up until the machine swaps, the fan spins up, or a tool starts timing out — and at that point it's hard to tell whether the slowdown is Claude Code, the OS, or something else.

I'd love a small, passive resource indicator rendered just below the input bar that lights up only when the local machine is under pressure.

Motivation

I use a lot of agents (Explore, code review, parallel subagents, background jobs) and over a long session I've seen:

• RAM creep up to the point macOS starts swapping
• CPU pinned by orphaned tool processes or MCP servers
• Disk pressure from large tool outputs, worktrees, or .claude/jobs/ debris
• Subtle UI/tool latency that I only correlate with system load after the fact

A quiet local hint would let me decide to /compact, kill a background job, exit a worktree, or restart the session before I hit a wall — rather than after.

A single-line, low-key status row directly under the input bar, only visible when a threshold is crossed. Examples:

[!] RAM 92% Claude Code using 4.1 GB · consider /compact or restart

Design notes:

• Passive, not modal — never blocks input, never steals focus
• Threshold-gated — hidden at normal load; only appears above e.g. 85% sustained
• Actionable — suggests the relevant command (/compact, /jobs, /worktree, etc.)
• Process-attributed when possible — distinguish "system is busy" from "Claude Code is busy"
• Configurable — settings.json knob for thresholds + an off switch for users who don't want it