Dispel System

• Outcome: Pass
• Latest audit: Jun 13, 2026 at 2:54 PM UTC
• Version:
  1.4.7

Overview

This is a markdown-only relationship-control analysis skill with sensitive but disclosed safety and recovery guidance, not hidden executable behavior.

Install only if you want a strong coercive-control analysis lens. If you are in immediate danger, may be monitored, or are considering documentation, recording, financial separation, legal action, or trauma exercises, prioritize real-world safety and consult local domestic-violence, legal, medical, or mental-health professionals rather than following the skill automatically.

SkillSpector

By NVIDIA

Vulnerability Patterns

• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Trigger Abuse: Overly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
• 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

Findings (11)

Vague Triggers

Medium

• Confidence: 89% confidence
• Finding: The trigger words are broad and include emotionally charged relationship-abuse terms, which can cause the skill to activate in ambiguous situations before a careful suitability determination is made. Although the document later includes a screening gate, the manifest-level triggers still increase the chance of over-invocation and premature framing of ordinary relationship distress as coercive control.

Missing User Warnings

Medium

• Confidence: 94% confidence
• Finding: The skill instructs users to collect screenshots, recordings, backups, and photos, but the warning about recording legality and evidence-preservation safety is incomplete and appears only inline with the action step. In abuse contexts, documentation can escalate danger if discovered, and recording may be illegal or unsafe in some jurisdictions, so giving operational collection advice without a stronger upfront risk framework can expose users to retaliation or legal harm.

Vague Triggers

Medium

• Confidence: 89% confidence
• Finding: The trigger phrases are expansive and semantically broad, covering common distress narratives without clear boundaries, exclusions, or required context. That can cause the skill to activate in situations outside its intended scope and deliver highly directive relationship-abuse framing to ambiguous cases, increasing the risk of misclassification and harmful guidance.

Missing User Warnings

Medium

• Confidence: 92% confidence
• Finding: The skill explicitly asks users to revisit dark periods, trauma-like events, coercive relationships, and high-distress memories, which can intensify distress, trigger flashbacks, or worsen dissociation. Although it includes a brief screening section, the protocol then proceeds into detailed memory excavation without robust guardrails, dynamic check-ins, or a clear crisis-routing fallback during the exercise itself.

Natural-Language Policy Violations

Medium

• Confidence: 95% confidence
• Finding: This section presents a speculative psychological model as universal fact, asserting that every person has an indestructible inner core that cannot be destroyed or taken away. In a vulnerable user context, such absolute claims can pressure users to conform to the framework, invalidate users who do not resonate with it, and encourage the agent to speak with unwarranted certainty about mental states and trauma recovery.

Natural-Language Policy Violations

Medium

• Confidence: 96% confidence
• Finding: The skill tells the agent to override user doubt with universal statements like 'everyone has this' and 'it has been there since birth,' which can coerce users into adopting a prescribed identity narrative. For users recovering from manipulation or coercive relationships, this is especially risky because it mirrors certainty-based reframing and may undermine autonomy, self-trust, or honest disagreement.

Vague Triggers

Medium

• Confidence: 88% confidence
• Finding: The trigger words are broad, symptom-based descriptions that can match many emotionally vulnerable situations without clear boundaries, exclusions, or scope constraints. In a relationship-analysis skill, this can cause over-invocation and funnel users in distress into a rigid interpretive frame that may misclassify normal conflict, reinforce paranoia, or crowd out safer escalation paths.

Vague Triggers

Medium

• Confidence: 91% confidence
• Finding: The trigger phrases are broad, emotionally charged, and map to common relationship-conflict language without clear boundaries, exclusions, or gating conditions. In a high-stakes interpersonal safety context, this can cause the skill to activate for ambiguous situations and deliver strongly pathologizing conclusions about real people, increasing the risk of misclassification, escalation, or unsafe advice framing.

Vague Triggers

Medium

• Confidence: 88% confidence
• Finding: The trigger phrases are broad, emotionally charged, and framed around vulnerability in relationships, which can cause the skill to activate for users in distress without clear gating, scope limits, or safety exclusions. In this context, overbroad activation is risky because the skill provides quasi-diagnostic interpretations and prescriptive framing about abuse/control dynamics, which may inappropriately shape user beliefs in ambiguous situations.

Vague Triggers

Medium

• Confidence: 90% confidence
• Finding: The trigger phrases are broad and map to common relationship confusion or support-seeking language, so this skill may activate for users who are not clearly in an abuse-analysis context. Because the content strongly frames situations through a control/abuse lens, accidental invocation can steer vulnerable users toward high-stakes interpretations without sufficient triage or context checks.

Missing User Warnings

Medium

• Confidence: 96% confidence
• Finding: This section gives concrete behavior-changing guidance for real-world abusive situations, including documenting, setting principles, and changing support dynamics, but does not prominently instruct users to prioritize immediate safety, emergency services, domestic-violence resources, or trained local support. In an abuse context, advice that encourages action without safety planning can escalate danger, especially if an abuser monitors communications or retaliates when control is challenged.

VirusTotal

64/64 vendors flagged this skill as clean.