Working-Set Overflow: When a Local Agent Should Yield to a Free Server

Описание и применение

При работе с большими локальными наборами данных агент может замедляться из‑за дисковой нагрузки и перегрева. С помощью классификации файлов, исключения секретов и проверки сети можно безопасно передать не‑секретную часть работы на бесплатный удалённый сервер, сохраняя при этом конфиденциальные данные локально.

Задача

A local coding agent stalls on large working sets, causing disk thrash and thermal throttling, while a free remote server can handle the bulk of the job but must avoid leaking secrets.

Результат

By staging only non‑secret files, probing connectivity, and classifying the secret surface, the engineer can safely offload bulky, non‑secret work to a free remote server while keeping secrets local, improving performance on resource‑constrained laptops.

Рабочий процесс

• 1. Pin the candidate tree by creating a staging directory that excludes dotenv files, private keys, and other secrets using rsync with --exclude patterns.
• 2. Classify the staged tree for secret surface by scanning for patterns such as .env, .pem, id_rsa, credentials.json, secret, and .kube/config.
• 3. Probe network connectivity with a short TCP probe to a trusted base URL; require consecutive successes before allowing a remote run.
• 4. If the secret surface is clear and the network probe succeeds, yield the non‑secret working set to the free remote server; otherwise, continue locally.
• 5. Monitor local latency metrics (page‑cache misses, thermal throttle) to validate that the remote run provides a performance benefit.

Доказательства

The script later times a stand‑in command that walks the working set the agent would touch.

Источник

Overflow never authorizes shipping dotenv files, private keys, session cookies, or customer extracts to any remote host.

Источник