Back to skill

Security audit

Kubernetes Service Mesh (Istio)

Security checks for vulnerabilities and agentic risk

Overview

This Istio management skill needs review because its production-traffic mirroring example lacks safeguards against duplicated actions and unintended sharing of request data.

Start in staging with namespace-limited permissions. Confirm the target cluster and intended changes, and save the current configuration for rollback. Before mirroring production traffic, ensure the destination is authorized to receive request data and cannot duplicate writes, payments, or notifications. Configuration checks and rollback examples help, but do not contain those downstream effects.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (5)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The supplied chunk targets Cilium and Hubble rather than Istio. Although network flow inspection could support troubleshooting, no Istio integration or supporting relationship is shown. The observable behavior therefore accesses resources inconsistent with the declared platform and does not implement the stated Istio management operations.

Content

No source excerpt is available for this finding.

Unqualified keywords such as "proxy" and "sidecar" can activate the skill outside an Istio context.

Medium
Category
Trigger Abuse
Confidence
94% confidence
Finding

The activation list includes generic keywords without requiring Kubernetes or Istio context. Requests about an HTTP proxy or another kind of sidecar could therefore invoke this mesh-management skill unintentionally.

Content

No source excerpt is available for this finding.

The workflow applies traffic-routing changes without warning about their effect on live requests.

Medium
Category
Output Handling
Confidence
96% confidence
Finding

The skill instructs the agent to execute kubectl_apply, rather than merely listing an available tool. This changes cluster routing and directs 10% of matching traffic to the canary, but the description provides no warning that live requests may be affected or fail if the canary is unhealthy.

Content

No source excerpt is available for this finding.

Production traffic mirroring is recommended without warning about duplicated request data or downstream side effects.

Medium
Category
Output Handling
Confidence
92% confidence
Finding

The example mirrors 100% of production requests to another service subset without restricting write operations or requiring isolated dependencies. Mirrored requests can still cause database writes, notifications, or other downstream effects even though their responses are discarded; copying production requests can also expose sensitive data to the canary. This is a legitimate testing technique, but presenting it without these safeguards makes the operational guidance unsafe.

Content

No source excerpt is available for this finding.

The workflow applies routing changes without warning that they can immediately affect live service traffic.

Medium
Category
Output Handling
Confidence
91% confidence
Finding

The workflow instructs the agent to apply new VirtualService weights, rather than merely listing an available command. It does not warn that applying the configuration changes routing in the selected cluster and namespace and may expose users to a failing service version; verification happens only after the change.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.