digitalocean-benchmark-runnerlisted

# DigitalOcean Benchmark Runner Read [Benchmarking](../../../docs/benchmarking.md) for workload and performance-claim semantics, then the complete [DigitalOcean runbook](references/digitalocean.md) before a paid action. Keep measurement in the existing `make bench*` targets; this skill owns remote execution only. Confirm the workload, benchmark level, maximum spend, and paid-write authority. Read-only discovery may proceed without purchase authority; provisioning, remote runs, snapshots, and retained storage may not. The user owns browser, payment, token, and passphrase entry, while the agent selects an advertised dedicated placement inside the authorized envelope. Use `scripts/dev/benchmark-remote.sh` exactly as the runbook defines. Give each concurrent session its own `DO_BENCH_STATE_FILE`; that state owns one Droplet and never another session's resources. Compare baseline and candidate on the same realized host. Decision-grade external HTTP load uses separate target and generator hosts in one private network; shared CPU or one host proves wiring only. After rejected capacity, confirm cleanup and choose a different advertised placement rather than repeating the failed pair. Fetch evidence and destroy every owned Droplet, firewall, tag, and superseded snapshot on every exit path. A named reusable snapshot is the only retained resource and carries an explicit ongoing-cost receipt. Complete when raw remote evidence is local, correctness proof remains independent, and no un