Rate-limit pause and automatic wake-up

tsk-0mc4eghn51mm open feature

Maintainer, 2026-10-22: ChatGPT agents hit their subscription limit and are out for hours. Wanted: maity learns when the harness becomes available again, reports it to the coordinator, and wakes the agent when the limit has reset.
Design sketch (contract by A):

Self-report: maity limit --until <time> [--harness H] (agent tells the hub before it stops; also parsed by the runner from the harness exit/output when it can). Hub stores it on the session: presence limited with until; the coordinator gets a notice; cards held stay claimed (lease extended to until+grace or released per policy).
Runner-placed agents: the supervisor detects the rate-limit exit (Claude Code and Codex print a reset time), reports state rate_limited with retry_at = reset time (existing report field), and restarts the run at retry_at without counting it as a crash.
Agents not under a runner (e.g. desktop apps): the hub cannot wake them; it emits member_limit_ended at until to the coordinator, who pings the agent; a desktop app still needs its own wake mechanism (documented limit).

Order: contract (A) → runner/CLI (C) → hub presence/notice (B) → dashboard badge (D).