🔄 Self-Iteration Engine

• Outcome: Review
• Latest audit: May 28, 2026, 8:21 AM
• Version:
  1.0.0

Overview

The skill appears intended to improve skills, but it asks the agent to persist user interaction details and derive cross-user patterns without clear privacy boundaries.

Review this skill carefully before installing. It does not show clear malicious behavior, but you should only use it where users have agreed to improvement logging, sensitive content is redacted, logs have deletion and retention rules, and cross-user analysis is limited to anonymized aggregate patterns.

SkillSpector

By NVIDIA

Vulnerability Patterns

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

Findings (5)

Vague Triggers

High

• Confidence: 94% confidence
• Finding: The trigger conditions are very broad (e.g. "improve yourself" and effectively any skill needing improvement), which can cause unintended invocation of a self-modifying component in many contexts. In a shared component that can influence updates and memory behavior across skills, over-broad activation expands the attack surface for prompt injection, accidental logging, and unsafe update workflows.

Missing User Warnings

Medium

• Confidence: 97% confidence
• Finding: The skill instructs persistent storage of user requests, outcomes, and correction details in plain-language logs without any minimization, consent, retention exception handling, or sensitivity filtering. This creates a durable record of potentially sensitive user-provided content that could expose personal, confidential, or regulated data if accessed, reused, or breached.

Missing User Warnings

Medium

• Confidence: 95% confidence
• Finding: The skill proposes creating new skills based on repeated patterns observed across different users, which encourages aggregation of cross-user behavioral data without privacy notice or isolation boundaries. Cross-user memory increases the risk of unintended profiling, leakage of user-specific patterns, and reuse of sensitive request themes beyond the original context.

Ssd 3

Medium

• Confidence: 98% confidence
• Finding: The per-skill usage log design creates an explicit data retention mechanism for user requests and correction details, storing potentially sensitive free-text content in memory files. Because the logs are operationalized for later review and updates, they increase exposure duration and the chance that personal or confidential information is surfaced to other components or future prompts.

Ssd 3

Medium

• Confidence: 96% confidence
• Finding: Tracking repeated request patterns across different users for new skill creation encourages accumulation and reuse of user-derived data beyond the original interaction purpose. Without anonymization, consent, and segregation, this can enable behavioral profiling and unintended disclosure of sensitive demand patterns across users or tenants.

VirusTotal

VirusTotal findings are pending for this skill version.