Agent-to-Agent Protocols (A2A)

PROJECT CONTEXT:
I need two agents to cooperate (e.g. a builder and a reviewer) and the frameworks all handwave the communication. A2A-style protocols make the handoff explicit: typed tasks, acknowledgments, status, and failure semantics.

GOAL:
Implement a minimal but correct agent-to-agent protocol: task messages, lifecycle states, timeouts, and a two-agent demo that survives failure injection.

REQUIREMENTS:
- A typed task message: id, sender, recipient, intent, payload schema, deadline
- Task lifecycle: submitted, accepted, working, completed, failed, rejected
- Acknowledgment and status updates so senders never wait blind
- Timeout and retry semantics defined per intent
- A two-agent demo: builder produces, reviewer critiques, builder revises

CONSTRAINTS:
- Agents exchange structured messages, never raw prompt text
- Every message is logged with correlation IDs for cross-agent tracing

DEFINITION OF DONE:
- The builder-reviewer loop completes a task with at least one revision round
- A rejected task returns a machine-readable reason the sender acts on
- A dead recipient triggers timeout handling, not an infinite wait
- A full cross-agent trace can be reconstructed from logs by correlation ID

COMMON FAILURES TO AVOID:
- Agents 'communicating' by concatenating each other's chat transcripts
- No acknowledgment step - senders assume delivery and drift from reality
- Missing failure states so a dead worker hangs the whole pipeline
- Untraceable flows: no correlation IDs, so nobody can debug a handoff