Chaos Engineer

• Outcome: Review
• Latest audit: May 29, 2026 at 9:23 PM UTC
• Version:
  0.1.0

Overview

This is a legitimate chaos-engineering skill, but several copyable examples can disrupt production systems or delete live resources without enough local safeguards.

Install only if you specifically need a skill for real chaos-engineering work. Treat its examples as hazardous templates: replace production targets with staging or isolated test resources, require explicit approvals, use least-privilege credentials, add dry-run and rollback steps, and avoid host-wide DNS edits outside disposable environments.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (7)

Missing User Warnings

Medium

• Confidence: 92% confidence
• Finding: The runbook contains real failure-injection commands that reboot an RDS instance with failover and revoke database network access, but there is no prominent safety warning immediately adjacent to those commands. In a chaos-engineering context these actions are expected, yet the lack of explicit guardrails increases the risk that a reader copies them into the wrong AWS account or environment, causing unintended outage or service degradation.

Missing User Warnings

Medium

• Confidence: 96% confidence
• Finding: This section includes guidance for terminating instances and deregistering targets to simulate AZ failure, but it does not prominently warn that these actions are intentionally destructive and can interrupt live services if run against production resources. In a chaos-engineering skill the actions are contextually expected, but the lack of blast-radius, environment, and approval guardrails makes accidental misuse materially risky.

Missing User Warnings

Medium

• Confidence: 98% confidence
• Finding: The resource exhaustion examples deliberately consume CPU, memory, disk I/O, and network bandwidth, but they do not clearly warn that they can degrade or fully interrupt host and network availability. Because these commands are copy-pastable and use aggressive defaults, an operator could easily cause denial of service on the wrong system.

Missing User Warnings

Medium

• Confidence: 98% confidence
• Finding: The Pumba examples kill, pause, impair networking for, and stop containers, including a pattern that could match production containers, without clear warnings about immediate service impact. In chaos engineering this is intentional behavior, but omitting explicit scoping guidance increases the chance of an accidental outage.

Missing User Warnings

Medium

• Confidence: 97% confidence
• Finding: The DNS chaos example modifies /etc/hosts with sudo, creating system-wide name resolution changes, but it does not clearly warn about privileged and persistent side effects. If cleanup fails or the process is interrupted, DNS behavior may remain altered and affect unrelated applications on the host.

Missing User Warnings

Medium

• Confidence: 93% confidence
• Finding: These examples perform intentionally disruptive actions against namespaces labeled as production, including pod failure, network partitioning, bandwidth throttling, disk latency injection, and recurring pod failure, but they do so without prominent safety warnings, staging-only guidance, rollback steps, or blast-radius constraints. In a chaos-engineering skill this behavior is contextually relevant, but the use of production targets and destructive defaults makes accidental misuse materially more likely, turning documentation into an unsafe operational recipe.

Missing User Warnings

High

• Confidence: 97% confidence
• Finding: The custom resource chaos script randomly deletes live custom resources in a production namespace and even includes an executable example invocation, yet provides no confirmation gate, dry-run mode, backup guidance, or warning that operator-managed resources may not be safely reconstructible. Because deletion can trigger irreversible service impact, state loss, or cascading operator actions, this is a dangerous pattern even in a chaos-engineering context.

VirusTotal

66/66 vendors flagged this skill as clean.