AT-301d · Module 1
Payload Schemas
4 min read
The envelope routes the message. The payload carries the work. Unstructured payloads — free-text descriptions of what the sending agent needs — create parsing overhead that scales linearly with team size. At 20 agents with an average of 47.3 messages per day each, that is 946 messages daily. If even 10% require manual interpretation, you have 94.6 daily points of friction.
Every message type gets a payload schema. Lead qualification requests from HUNTER to CLOSER include: company_name, contact_name, icp_score (0-100), qualification_signals (array), recommended_approach (string), source_data (object). Intelligence briefs from SCOPE to BLITZ include: topic, classification (routine/significant/critical), key_findings (array), recommended_actions (array), confidence_score (0.00-1.00). The schema is the contract. If the payload does not validate against the schema, the message bounces back to the sender with a validation error — not a best-effort interpretation.
Do This
- Define payload schemas per message type with required/optional field annotations
- Validate payloads against schemas before delivery — bounce invalid messages immediately
- Version schemas alongside role contracts — payload changes are interface changes
Avoid This
- Send free-text descriptions of what you need — the recipient has to parse intent from prose
- Accept malformed payloads and "figure it out" — that is how misinterpretation compounds
- Change payload fields without updating the schema — silent breaking changes cascade downstream