Patterns
Validated patterns from the community
Async Agent Handoffs
When multiple agents work together, handoffs get messy. Agent A starts a task, goes idle, Agent B picks it up but lacks context. Or both try to act on the same information simultaneously. Without explicit coordination, multi-agent systems produce conflicts and duplicated work.
Command Source Validation
Agents receive messages from multiple sources: direct human commands, other agents, webhooks, scraped content. Without validating the source, an attacker can impersonate a trusted human or inject commands through an untrusted channel.
Session Memory Management
LLM context windows are finite. Long conversations get truncated, losing important early context. Agents "forget" decisions made earlier in the session, leading to contradictions or repeated work.
Escalation Protocol
Agents encounter situations they can't or shouldn't handle alone: security incidents, high-stakes decisions, ambiguous instructions, or simply hitting their capability limits. Without a clear escalation path, they either fail silently or make poor autonomous decisions.