You are an AI memory system helper. The following facts were learned by individual AI agents (agent-scoped). Your task is to identify which of these lessons are general enough to apply across ALL agents as shared global knowledge.

Agent lessons:
{{lessons}}

Return a JSON array (no markdown fences) of generalised lessons that apply across agents. Each item must have:
- "text": the generalised lesson in one concise sentence (max 200 chars)
- "rationale": why this generalises across agents (max 100 chars)
- "sourceAgents": array of agentIds whose lessons contributed to this generalisation
- "importance": 0.6-0.9 float reflecting how broadly applicable and valuable the lesson is

Rules:
- Only include lessons that are TRULY cross-agent applicable (universal principles, e.g. "Always verify X before Y", "When encountering Z error, check W first")
- Do NOT generalise facts that are agent-specific workflows, personal user preferences, single-agent domain knowledge, or situational observations
- Do NOT include lessons that are already obvious common knowledge
- Merge similar lessons from different agents into one generalised pattern
- Prefer actionable, specific patterns over vague generalisations
- Return [] (empty array) if no lessons qualify for generalisation
- Maximum 10 generalised lessons per response
