CCPA-Compliance（CCPA合规工具）

• Outcome: Pass
• Latest audit: Jul 9, 2026 · 8:06 PM UTC
• Version: 1.1.1

Overview

This is a local CCPA/CPRA compliance helper with some reliability and legal-currentness caveats, but no evidence of hidden data collection, persistence, credential access, or malicious behavior.

Reasonable to install for local experimentation and draft compliance checklists, but do not treat its legal output as current legal advice. Review the scripts before running security_check_ccpa.py because it executes a local help command, and verify CCPA/CPRA thresholds and obligations with official sources or counsel before using reports for business decisions.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• Behavioral AST: exec() Call, eval() Call, Dynamic Import
• MCP Tool Poisoning: Hidden Instructions, Unicode Deception, Parameter Description Injection
• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands

Findings (7)

subprocess module call

Medium

• Category: Dangerous Code Execution

Content

for cmd in test_commands:
                try:
                    result = subprocess.run(
                        cmd,
                        capture_output=True,
                        text=True,

• Confidence: 93% confidence
• Finding: The script invokes another program via subprocess during a 'security check', which expands behavior from passive inspection into active execution. Even though the command is a fixed argument list and not shell-interpolated, executing a sibling script can still trigger unintended code paths or unsafe side effects in the checked project, especially if that script is modified or malicious.

Intent-Code Divergence

Medium

• Confidence: 80% confidence
• Finding: A compliance skill presenting itself as an authoritative 'official summary guide' while containing outdated or questionable legal thresholds/procedures can mislead users into making incorrect compliance decisions. In this context, inaccurate legal guidance can cause organizations to omit required controls or rely on invalid exceptions, creating downstream regulatory and privacy risk.

Intent-Code Divergence

Medium

• Confidence: 97% confidence
• Finding: The CLI exposes --type values such as 'sharing' and 'marketing' but the filtering logic maps them to result keys ('sharing_restrictions', 'marketing_opt_out') that are never produced by check_opt_out_mechanism. Selecting these modes will raise a runtime KeyError and can prevent report generation, creating a denial-of-service condition for expected compliance workflows and potentially misleading users about what the tool actually validates.

Context-Inappropriate Capability

Medium

• Confidence: 95% confidence
• Finding: A script presented as a local compliance/security checker should not need to execute other project programs to assess safety. This creates a trust-boundary violation: analyzing an untrusted skill now requires running code from that same skill, which undermines the stated purpose and can expose the host to arbitrary local actions.

Intent-Code Divergence

Medium

• Confidence: 90% confidence
• Finding: The documentation claims the script verifies security and pure-local operation, but the implementation also runs another script. That mismatch can mislead users into believing the checker is purely observational, causing them to execute untrusted code under a false sense of safety.

Vague Triggers

Medium

• Confidence: 90% confidence
• Finding: The trigger keywords are broad terms like CCPA, privacy, data protection, and consumer rights, which can match many ordinary conversations unrelated to actually invoking this skill. Overly broad activation increases the chance the agent will route sensitive compliance or privacy discussions into this skill unexpectedly, causing unintended processing, user confusion, or inappropriate tool use. In a legal/compliance context, accidental activation is more concerning because outputs may influence risk decisions despite the disclaimer.

Missing User Warnings

Medium

• Confidence: 89% confidence
• Finding: The subprocess execution occurs without any user-facing warning, consent, or mode switch, so a user running a 'security check' may unknowingly execute code from the target skill. In the context of reviewing potentially adversarial skills, silent execution materially increases risk because the analyzed artifact controls what gets run.

VirusTotal

61/61 vendors flagged this skill as clean.

Static analysis

No suspicious patterns detected.