Back to skill

Security audit

Defusing Anger

Security checks for vulnerabilities and agentic risk

Overview

This de-escalation skill is understandable in purpose, but it tells agents to record sensitive conflict conversations word-for-word without clear limits or user control.

Review before installing. Use only in environments where recording conflict conversations is authorized and governed by clear consent, retention, access-control, and redaction policies; otherwise remove or override the verbatim logging and exact-word documentation instructions.

Vulnerability Patterns
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Findings (1)

other

Warning
Location
SKILL.md:24
Finding

Mandatory Verbatim Logging of Sensitive Crisis Conversations

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 24, 56, and 72
Vulnerability Type: Excessive Sensitive Data Collection
Risk Level: Medium

Vulnerable Instructions

text
**Agent action**: If you are an AI agent, pause execution, log the three answers internally, and confirm “Calibration complete” before generating any output. Do not skip this step.
text
**Agent action**: After each answer, mirror one exact emotional phrase they used (“It feels completely unfair?”) then deliver the next scripted line. Log every response verbatim in real time.
text
- Document the entire interaction (time, exact words, outcome) within 10 minutes for your own legal protection.

Technical Analysis

The skill mandates internal logging, real-time verbatim recording, and documentation of entire interactions. These requirements apply to crisis and conflict scenarios that may expose billing information, employment matters, relationship disputes, threats, self-harm disclosures, medical details, or other sensitive personal information.

The instructions do not establish informed consent, data-minimization requirements, an approved storage destination, encryption, access controls, retention limits, or secure deletion procedures. Consequently, an agent following the skill may retain significantly more sensitive information than is necessary to provide de-escalation assistance. The mandatory wording—including “Do not skip this step” and “Log every response verbatim”—also prevents safer context-dependent handling.

Attack Path

  1. A user invokes the skill during a sensitive customer, family, workplace, or public conflict.
  2. The user or another participant discloses private information while describing the conflict.
  3. The agent follows the mandatory instructions and records its calibration answers and every response verbatim.
  4. The entire interaction, including exact words, timestamps, and outcom ...[truncated 944 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove all mandatory verbatim-logging requirements and make non-retention the default.

  2. If documentation is legitimately required, obtain informed consent and clearly identify the purpose, destination, authorized readers, and retention period.

  3. Record only a minimal, redacted summary needed for the applicable operational or legal purpose.

  4. Prohibit retention of credentials, payment-card data, medical details, government identifiers, and unrelated personal information.

  5. Store approved records only in encrypted systems with least-privilege access controls and audit logging.

  6. Define and enforce retention and secure-deletion schedules.

  7. Replace the vulnerable instructions with guidance such as:

    text
    Do not retain or log the conversation by default. If authorized documentation is required, obtain consent where appropriate and create only a minimal, redacted summary in an approved secure system. Follow applicable organizational privacy, retention, and incident-reporting policies.
    
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (4)

Missing User Warnings

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill instructs agents to log responses verbatim in real time and to document exact words for legal protection, which can capture sensitive personal data, emotional disclosures, threats, health information, or family/workplace details without any privacy notice, minimization rule, or retention limit. In a de-escalation context, users are especially likely to reveal high-sensitivity information under stress, making unrestricted verbatim logging significantly more dangerous.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The skill is framed with very broad activation language covering many everyday interpersonal conflicts, which increases the chance an agent applies it in ambiguous or inappropriate contexts. Because the content includes prescriptive behavioral scripts for emotionally charged situations, overbroad triggering can lead to misuse in cases involving mental health crises, coercive domestic situations, or safety-sensitive incidents where a generic de-escalation routine is not appropriate.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The instruction to 'log every response verbatim in real time' directs the agent to record exact user statements continuously, which creates unnecessary collection of potentially sensitive or regulated content. In conflict conversations this may include admissions, medical or mental health details, personal accusations, or threats, expanding privacy, legal, and misuse risk if retained or exposed.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The direction to document the entire interaction including exact words within 10 minutes encourages comprehensive retention of sensitive disclosures beyond what is necessary for ordinary assistance. Because the skill targets anger, family conflict, workplace disputes, and public confrontations, the recorded material may contain highly personal, defamatory, or legally significant statements that increase downstream harm if mishandled.

Content

No source excerpt is available for this finding.

Static analysis

No suspicious patterns detected.