digitalocean-benchmark-runner

# DigitalOcean Benchmark Runner Read [Benchmarking](../../../docs/benchmarking.md) and the [DigitalOcean runbook](references/digitalocean.md). Keep measurement in the repository's existing `make bench*` targets; this skill owns only remote execution. ## Execute 1. Confirm the workload, maximum spend, benchmark level, and whether external paid writes are authorized. `check`, `list`, and `image-list` are read-only; `create`, `run`, and `image-build` incur cost, while a created snapshot keeps incurring storage cost until deletion. One authorization covers agent-selected placements and bounded retries that stay inside its cost, security, and proof envelope; do not seek approval per host or attempt. 2. For an ordinary benchmark request, use DigitalOcean only when `doctl` is already installed and the selected context is authorized. Otherwise stop the remote path and use the matching local command. When the user explicitly asks to configure DigitalOcean, follow the runbook's account/workstation onboarding: the user owns browser, payment, token, and passphrase entry; the agent may perform the requested local setup and read-only verification but must never request or echo those secrets. A real setup smoke still needs explicit paid-write authorization. 3. Run `scripts/dev/benchmark-remote.sh list` and resolve placement from current provider data before provisioning. `DO_BENCH_REGION` and `DO_BENCH_SIZE` are agent-owned inputs unless geography or hardware is part of the accepted clai...