Execution at runtime

How EarnClaw runs agent cycles, and how Thirdfy powers governed execution for hosted runtimes.

When a finance agent runs a live cycle, EarnClaw applies the policy you configured in the wizard before any onchain action executes. You see the outcome in run history: executed intents or a documented skip with a reason code.

Configure delegation and guardrails in Create your first agent. Manage your runtime plan in Settings. See Runtime plans and execution and Fund and run.


What runs on each cycle

LayerWhat it does in EarnClaw
PreflightChecks the intent against org policy, delegation, template rules, and funding
DelegationLimits what the agent wallet may sign, within scopes and expiry you set
Intent executionRuns approved onchain and tool-backed steps when preflight passes
Execution allowanceMeters usage against your runtime plan monthly budget per runtime

Thirdfy partnership

EarnClaw partners with Thirdfy for governed execution. You stay in the EarnClaw workspace. Thirdfy provides the execution layer behind hosted agents.

RuntimeHow it connects
HermesFly-hosted packs use Thirdfy preflight and execute-intent through the Agent CLI and API
OpenClawSame execution rails; OpenClaw runtimes call Thirdfy for catalog, delegation, and wallet actions
Claude managedThirdfy MCP connects managed sessions to the same policy and execution surface

Execution volume is included in your EarnClaw runtime plan. You do not set up a separate Thirdfy retail account to deploy on EarnClaw.

For protocol depth (CLI, MCP, API, delegation, venues, and the credits model), use Thirdfy Docs:


Failed preflight is normal

When policy, funding, delegation, allowance, or venue readiness blocks an action, the agent skips and logs why. That is expected behavior, not a broken agent. Repeated skips with the same reason usually mean a setup gap: wallet funding, expired delegation, plan allowance exhausted, or inactive subscription.

See Observability and Schedules and guardrails.