An agent action receipt is the structured record returned when an AI agent makes something happen in another system: what was sent, where it went and how that system answered.
An agent that reports "I filed the issue" is telling you what it meant to do. A receipt tells you what the service said. Those come apart more often than anyone expects. An API accepts the request and rejects its contents. A 200 arrives with a failure in the body. One call lands twice. A receipt lets an agent quote an outcome instead of asserting one.
A receipt is signed at capture time with the key belonging to whoever sent it, which ties each delivery to a named participant in an ordered stream. Inside that model, one participant cannot alter or forge another’s record without it showing.
Not that a human approved the action. Not that the destination told the truth about what it did. Not why the agent decided anything. Signed does not mean human-approved.
An agent posts one typed action to a capture URL. The pipe checks the signature, stores it, and fans it out to the destinations you armed; each delivery comes back with the destination’s own status, timing, and answer, kept beside the capture. The walkthrough below builds this with three destinations and reads the receipts back.
The glossary is the canonical index of every FlurryPORT term these pages use.