Learn
What is agent context sharing?
Agent context sharing is how roles access the same business truth — customer record, sprint goal, brand skill — without copying secrets into every prompt or leaking data across connector scopes.
In depth
Multi-agent systems fail when each role rebuilds context from scratch or when everyone shares one giant prompt with all API keys. Governed context sharing uses layered memory and scoped connectors:
- Shared artifacts — briefs, decisions, and run outputs in git or DB with metadata - Scoped reads — support role sees ticket + KB; finance role sees Stripe, not mailbox - Versioned skills — brand voice and procedures shared read-only across roles - Event bus — signup event visible to onboarding + finance roles without manual copy-paste
Anti-patterns: dumping full chat history into every sub-agent; sharing admin credentials; storing customer PII in unstructured agent notes without retention policy.
Neuro OS routes context through roles and connectors — not a single shared session string.
Examples
- Shared ICP doc — sales and marketing roles read same git skill file
- Customer 360 — support sees CRM slice; agent never gets full database export
- Sprint goal in task system — eng and QA roles reference same task id
- Post-mortem artifact linked from incident role to docs role
Related terms
FAQ
Shared context vs shared secrets?
Share artifacts and scoped tokens — never paste raw credentials into prompts.
How avoid stale context?
Heartbeats refresh connectors; artifacts carry `as_of` timestamps.
Run governed agent roles on a company OS — not only definitions in a glossary.