Safety-Posture Decision
- Entity ID:
ent-20260423-d004e1000002 - Type:
decision - Scope:
private - Status:
active
Description
One of six recurring design decisions. Options: deny-first layered enforcement (Claude Code, 7 layers), container isolation (SWE-Agent/OpenHands), VCS rollback (Aider), or approval-only. Key guide insight: defense-in-depth only works if layers have INDEPENDENT failure modes; Claude Code's layers share an economic constraint (commands >50 subcommands bypass security analysis entirely).
Key claims
- Defense-in-depth only works when safety layers have independent failure modes
Relations
- Safety-Posture Decision --[answered_by]--> Deny-First with Human Escalation (Principle)
- Safety-Posture Decision --[answered_by]--> Defense in Depth with Layered Mechanisms (Principle)