cpu-gpu-performance

• Outcome: Review
• Latest audit: Sep 16, 2026 · 5:59 AM UTC
• Version: 1.9.19

Overview

The skill is mostly a resource-monitoring checklist, but it asks agents to collect full local process command lines in a way that can expose secrets unnecessarily.

Review this skill before installing in environments where process arguments may contain tokens, passwords, database URLs, customer names, or private file paths. If used, replace the process command with a sanitized CPU-sorted form such as ps -eo pid,comm,pcpu --sort=-pcpu | head and avoid copying raw process arguments into notes or final reports.

A.I.G

Vulnerability Patterns

• Insecure Skill Coding Practices: Finds exploitable flaws such as hardcoded secrets or command injection
• Skill Instruction Hijacking: Alters the agent's session goals or safety constraints when the skill loads
• Agent Memory Poisoning: Writes attacker-controlled rules into memory that affect later sessions
• Remote Payload Retrieval and Execution: Fetches external code whose behavior can change after review
• Embedded Malicious Code: Ships malicious scripts inside the skill and executes them locally

Findings (1)

T09 · Insecure Skill Coding Practices

Warning

• Location: SKILL.md:45

Finding

Full Process Command-Line Collection May Expose Sensitive Arguments

Content

View full analysis

Vulnerability Details

File Location: SKILL.md, lines 45-52; related logging instructions at lines 88-91
Vulnerability Type: Sensitive information exposure through insecure process inspection
Risk Level: Medium

Vulnerable Code Snippet:

markdown

## Step 1: Establish Current Baseline
- Capture current utilization:
  - `uptime`
  - `ps -eo pcpu,cmd | head`
  - `nvidia-smi --query-gpu=utilization.gpu,memory.used --format=csv`

  Note which hosts/GPUs are already busy.
- Record any CI/cluster budgets (time quotas, GPU hours) before launching work.
- Set a per-task CPU minute / GPU minute budget that respects those limits.

The collected information is subsequently subject to documentation requirements:

markdown

Conclude by documenting the commands that were run and their resource cost
(duration, CPU%, GPU%), confirming whether they remained within the per-task
budget.

Technical Analysis

The command ps -eo pcpu,cmd | head requests complete process command lines through the cmd field. Process arguments can contain API tokens, passwords, database connection strings, confidential file paths, customer identifiers, or other sensitive job parameters.

Complete command lines are not necessary to establish a CPU-utilization baseline. Moreover, the command does not specify CPU-based sorting, so head returns the first process-table entries rather than reliably selecting the processes with the highest CPU usage. This creates avoidable information exposure without consistently fulfilling the stated monitoring objective.

The skill also directs the agent to document commands and resource observations. If sensitive process arguments enter the agent context, they may subsequently be copied into notes, logs, conversation output, or another reporting system.

Attack Path

1. A process visible to the user running the skill is launched with sensitive data in its command-line arguments ...[truncated 1254 chars]

Remediation

View remediation

Remediation Suggestions

1. Avoid collecting full process arguments. Replace the command with a sanitized, CPU-sorted view:

   sh

   ps -eo pid,comm,pcpu --sort=-pcpu | head
   

2. Use comm rather than cmd or args so only executable names are collected.
3. Explicitly prohibit copying process arguments, credentials, tokens, connection strings, or other secrets into notes and reports.
4. Apply redaction before retaining any monitoring output, including filtering common secret-bearing flags and environment-derived values.
5. Limit baseline collection to sessions that will perform resource-intensive work rather than automatically collecting process information at every session.
6. Document only aggregate utilization metrics and sanitized process identifiers required for resource planning.

SkillSpector

Vulnerability Patterns

• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code
• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep

Static analysis

No suspicious patterns detected.