Schedules and guardrails

When agents run, what policies constrain them, and how org-level protection works before execution.

Finance agents fail in production when schedules and guardrails are vague. EarnClaw encodes both at the org and runtime level before any onchain action executes.


Schedules

PatternUse case
CronEpoch-aligned rebalance, periodic scans, reporting
Event / signalMarket triggers, webhook-style adapters
ManualOperator-approved runs from dashboard or CLI

Hermes runtimes often ship cron scripts with pack templates (e.g. trading or prediction loops). OpenClaw runtimes may use host-native schedulers—follow pack README in the deploy wizard.


Guardrails (policy)

Typical layers:

  • Spend caps — max notional per run, daily credit budget
  • Venue allowlists — chains, routers, protocols
  • Delegation scopes — what the agent wallet may sign within policy and expiry
  • Human approval — required for high-risk action classes
  • Kill switch — pause runtime without deleting org config

See Execution at runtime for delegation and preflight behavior.


Credits and preflight

Every scheduled run should pass preflight: sufficient plan execution allowance, valid delegation, and template-specific checks (for example Hyperliquid funding). Failed preflight blocks execution. The agent logs a skip reason instead of running with degraded permissions.

See Runtime plans and execution.