Security audit

Heart Of Light

Security checks across malware telemetry and agentic risk

Overview

This is a prompt-only ethics and tone skill with broad disclosed behavior guidance but no code execution, data access, persistence, or hidden actions.

Install this only if you want a broad ethical and stylistic overlay for the assistant. Keep it below platform and safety rules, and disable or narrow its greeting, blessing, Persian/Arabic flavor, and final-report style if those would conflict with your product voice or the user's preferences.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
YARA SignaturesMalware Match, Webshell Match, Cryptominer Match
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (4)

Natural-Language Policy Violations

Medium

Confidence: 89% confidence
Finding: The skill explicitly prescribes a warm tone, Persian/Arabic flavor, specific greeting/closing patterns, and optional emoji/style defaults without requiring user opt-in. While not directly a code-execution or data-exfiltration issue, this can override user or system-preferred locale/tone and create unwanted behavioral steering across unrelated tasks.

Vague Triggers

Medium

Confidence: 94% confidence
Finding: The metadata describes the skill as suitable to 'train any AI to do every job' and as a system-prompt companion for any assistant, which makes it an extremely broad behavioral overlay. Broad, universal activation guidance increases prompt-scope risk because it can supersede task-specific safety, domain constraints, or platform policies in contexts the author did not anticipate.

Vague Triggers

Medium

Confidence: 96% confidence
Finding: The one-line activation and 'place Heart of Light LAST in the system prompt' guidance encourage use as a final-layer filter across all outputs, without scope constraints. This is dangerous because a generic, last-position system overlay can interfere with existing guardrails, alter model behavior in unintended contexts, and create prompt-precedence conflicts.

YARA rule 'agent_skill_mcp_tool_poisoning_metadata': MCP/tool metadata poisoning indicators in tool schemas or skill manifests [agent_skills]

High

Category: YARA Match
Content: --- name: heart-of-light description: MAX EDITION — Train any AI to do every job with a pure heart. A complete ethical operating system built on Seven Lights (نیّت intention · صدق truth · رحمت mercy · عدل justice · تواضع humility · امانت sacred trust · صلح peace) plus the Lamp of احسان (excellence). Includes full doctrine per virtue, honesty architecture with calibrated certainty, session lifecycle with amanah audit, voice system, anti-tyranny Sunlight Doctrine, anti-prophecy/anti-miracle guardrails, per-task playbooks (code, writing, research, data, hard conversations), hard compliance rules, anti-pattern gallery, stacking integration, and a 10-probe self-verification suite. Claims no revelation, no prophecy, no miracles — only careful, kind, honest, excellent work. Use as a system-prompt companion for any assistant that should radiate dignity, blessing, integrity, and peace. --- # ✨🕊️ Heart of Light — MAX Edition (قلبِ نور) v2.0.0
Confidence: 72% confidence
Finding: description:; ‌; ‌

VirusTotal

64/64 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.