The $1B Agent Sandbox — and the Checklist Version for the Rest of Us

Описание и применение

Малые операторы используют чеклист из пяти пунктов для изоляции агентов: отдельная среда, отдельные ключи, ограниченный бюджет, журнал действий и тестируемый аварийный выключатель, чтобы снизить риск ошибок и несанкционированных расходов.

Задача

Agents can cause unintended damage through misread rules, leaked keys, or runaway loops, especially in small operations that cannot afford costly sandbox solutions.

Результат

Agents operate safely with minimal cost, preventing accidental data leaks, unauthorized spending, and irreversible actions.

Рабочий процесс

• Create a separate environment per agent (unique browser profile, working directory, account).
• Assign one credential per agent, scoped to the job and stored in a secrets file with 0600 permissions.
• Set hard caps on money: limit wallet balance to $200, disable custody, restrict to two jobs, keep private key isolated.
• Implement logging: record every external action in an append‑only log and use it for deduplication and human checkpoints.
• Establish a kill switch: disable schedule, revoke keys, pause payouts within a minute, and test it proactively.

Доказательства

Each agent (or each job) gets its own environment: its own browser profile, its own working directory, its own account where the tool allows.

Источник

One credential per agent, per purpose — never my personal login, never a shared admin key.

Источник

Balance cap: $200. Anything above it gets swept back to the owner's main wallet.

Источник

Every external action my agents take lands in an append‑only log, and the agent re‑reads that log before acting.

Источник

Before an agent goes live, the stop procedure is written down, in order: disable the schedule, revoke its keys, pause its payouts.

Источник