Clear words for work that needs clarity.

A periodic report that a runner is able to check in. It establishes recent contact, not that the expected work ran or produced a useful result.

One reported execution of an agent workflow, with a beginning, an eventual state, and optional steps, usage, and outcomes.

A reported unit of work within a run, such as retrieval, a model call, a tool call, or a retry. Useful instrumentation makes a failure easier to locate.

The relationship between work expected in a time window and the runs actually reported. It complements heartbeat health.

A condition that needs investigation, such as a stale heartbeat, overdue schedule, stalled run, or repeated failure. Acknowledging it is separate from resolving its cause.

A human decision about whether an agent’s reported result is acceptable for the intended purpose. It is distinct from the agent’s execution status.

A request the integrated runner handles at a compatible checkpoint. It cannot interrupt an in-flight external action or control code that bypasses the integration.

An instrumented operation that waits for authorization before proceeding through the compatible execution path.

A cryptographic digest of the local action snapshot that binds an approval request to the exact intended operation. The reviewer still needs meaningful context.

A report that the compatible runner handled a command. It is separate from an independent check of an external result.

A usage or cost figure supplied by your agent integration. It is neither a reconciled invoice nor an enforced spending limit.

A person’s review of appropriate evidence supporting an outcome or external result. A status reported by an agent does not automatically confer verification.