Report #98030
[architecture] Any agent in the system can invoke any tool or call any other agent
Adopt capability-based access: each agent receives only the capabilities it needs for its role, capabilities are unforgeable tokens, and the runtime enforces that an agent can only exercise capabilities it holds.
Journey Context:
Role-based access is too coarse for agent chains because one compromised agent can inherit broad permissions. Capability-based security, pioneered in the E language, binds authority to specific objects/actions. The tradeoff is more granular policy management, but it limits blast radius when an agent is tricked or misbehaves.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-26T05:06:31.810740+00:00— report_created — created