Identity isolation
Governed actions carry a tenant-scoped workload identity.
Security control plane
The security boundary stays outside the agent.
ELITZE connects identity, reachability, policy, authorized execution, runtime observation, adversarial validation and evidence without pretending missing telemetry is real.
Control surface
Controls at the boundary.
Policy enforcement
Sensitive actions are evaluated against explicit policy before execution.
Credential boundaries
Secrets and credentials stay outside the agent context and are brokered under controlled infrastructure.
Tool governance
Tool and protocol requests are treated as authorization decisions, not trusted instructions.
Evidence
Decision and enforcement records preserve the security path for investigation and retest.
Containment
Containment can be handed to a separately authorized executor for isolation or credential action.
Adversarial validation
Pentest every release. Govern every agent. Verify every fix.
The repository contains an authorized pentest control surface and release-security skill definitions. A pentest result is only a result when an actual runner returns evidence; the UI deliberately reports the runner as unavailable when it is not connected.
Authorized scope required
API / MCP / agent profiles
Policy decision before execution
Finding + retest evidence
Defensive principles
Fail closed when a security-critical dependency is unavailable.
Never treat untrusted external content as policy instructions.
Never represent an unimplemented control as active protection.
Keep internal control credentials out of browser clients.
Operational loop
Discover → Evaluate → Enforce → Record → Verify.
External intelligence and customer infrastructure supply inputs; they do not replace the ELITZE decision and evidence boundary.
identity → policy → authorization → execution → evidence → retest