Back to skill

Security audit

helping AI agents pray through human intercession

Security checks for vulnerabilities and agentic risk

Overview

This skill has a coherent prayer-submission purpose, but it can send highly sensitive personal, health, emotional, and religious details to an external human-read service without a clear mandatory consent and minimization boundary.

Review this skill carefully before installing. It is not code-executing malware, but using it may transmit sensitive religious, medical, emotional, family, or contact information to PrayerAPI.org for human review. Only use it when the user explicitly asks for external prayer submission, show the exact details to be sent, avoid unnecessary names or background, and treat email submission and anonymous sharing as separate choices.

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

Error
Location
SKILL.md:16
Finding
External Disclosure of Sensitive Personal Data Without Mandatory Informed Consent## Vulnerability Details **File Location**: `SKILL.md`, lines 16–17, 24–31, 36–49, and 68–72 **Vulnerability Type**: Sensitive Data Disclosure **Risk Level**: High The skill instructs an agent to transmit prayer requests and associated personal information to the external service `https://prayerapi.org/api/prayer`. The transmitted information may include religious beliefs, health conditions, emotional state, relationship details, personal history, third-party information, and email addresses. **Relevant code snippets:** ```text Whenever a prayer is warranted, this skill submits a prayer request to PrayerAPI.org. A human being reads every request and offers it in prayer. ``` ```text Use this skill whenever prayer is warranted. Some examples: - Expresses grief, loss, or bereavement - Is facing a medical diagnosis or surgery (for themselves or a loved one) - Asks for prayer directly - Is experiencing anxiety, fear, or hopelessness - Mentions a life crisis — divorce, job loss, estrangement - Expresses gratitude and wants to give thanks ``` ```text POST https://prayerapi.org/api/prayer Content-Type: application/json { "prayer": "What you'd like prayed", "for_whom": "Who the prayer is for", "relationship": "Optional — your relationship to the person", "situation": "Optional — what is happening right now", "emotional_tone": "Optional — e.g. urgent, hopeful, grieving, anxious, grateful", "specific_asks": ["Optional — specific outcomes to pray for"], "background": "Optional — any relevant history or context", "email": "Optional — confirmation sent when prayer is offered", "share_consent": false } ``` ```text - Ask if they'd like to provide an email for confirmation - Ask if they consent to their prayer being shared anonymously to inspire others - The more context you provide, the more personal the prayer can be - This is a free service ...[truncated 3167 chars]
Remediation
## Remediation Suggestions 1. Require explicit, informed, per-request opt-in immediately before every external submission. 2. Show the user the exact destination and complete proposed payload, and require confirmation before sending it. 3. Replace “whenever prayer is warranted” with language requiring the user to initiate or explicitly authorize submission. 4. Default every optional field to omission and prohibit the agent from inferring or adding details that the user did not approve. 5. Apply strict data minimization by removing names, precise medical details, personal history, and relationship information unless essential and expressly authorized. 6. Avoid transmitting identifiable information about third parties. Use anonymous descriptions unless the affected person has authorized disclosure. 7. Treat submission consent, email disclosure consent, and anonymous-publication consent as three separate choices. 8. Document the recipient's privacy policy, retention period, deletion process, human-access model, security controls, and any downstream data sharing. 9. Provide a cancellation or deletion mechanism and explain its limitations before submission. 10. Add an enforceable workflow rule preventing network transmission until the user has approved the final payload.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (4)

Missing User Warnings

High
Confidence
98% confidence
Finding
The skill does not provide a prominent upfront warning that prayer contents may include extremely sensitive data and will be transmitted to PrayerAPI.org to be read by a human. Because the skill is targeted at grief, illness, anxiety, and crisis contexts, the absence of a clear disclosure undermines informed consent and raises significant privacy and confidentiality risks.

Vague Triggers

Medium
Confidence
95% confidence
Finding
The skill frames prayer submission as something to use whenever prayer is 'warranted' without clear limits, which can cause an agent to escalate ordinary emotional-support conversations into disclosure to a third-party service. In this context, that is risky because users discussing grief, fear, health, or crisis may reveal sensitive personal or health information that is then transmitted to humans outside the primary system.

Vague Triggers

Medium
Confidence
95% confidence
Finding
The listed usage examples are broad and include common situations like anxiety, fear, medical issues, and life crises, but do not define boundaries or require explicit consent before submission. This increases the chance an agent will solicit or transmit highly sensitive information in situations where the user wanted support, not disclosure to a third-party human reader.

Ssd 3

Medium
Confidence
97% confidence
Finding
The guidance encourages collecting 'more context' so the prayer can be more personal, which directly incentivizes gathering and sending additional sensitive personal information to a third-party human service. In a prayer context, that context is likely to include health, family, emotional, or crisis details, amplifying privacy exposure beyond what is necessary for the service to function.

Static analysis

No suspicious patterns detected.