Improve Relationships

• Outcome: Pass
• Latest audit: Sep 11, 2026 · 8:03 AM UTC
• Version: 0.2.2

Overview

This skill uses sensitive relationship context to suggest and draft outreach, but its behavior is disclosed, draft-only, and aligned with its stated purpose.

Install this only if you are comfortable letting the agent analyze relationship context you provide or connect, such as calendars, messages, contact notes, and shared social pages. Review every draft carefully, especially inferred language, cultural or religious occasions, and personal details, before sending anything yourself.

A.I.G

Vulnerability Patterns

• Skill Instruction Hijacking: Alters the agent's session goals or safety constraints when the skill loads
• Agent Memory Poisoning: Writes attacker-controlled rules into memory that affect later sessions
• Remote Payload Retrieval and Execution: Fetches external code whose behavior can change after review
• Embedded Malicious Code: Ships malicious scripts inside the skill and executes them locally
• Unauthorized Access and Privilege Escalation: Obtains permissions beyond the task's legitimate needs

SkillSpector

Vulnerability Patterns

• Prompt Injection: Instruction Override, Hidden Instructions, Exfiltration Commands
• Data Exfiltration: External Transmission, Env Variable Harvesting, File System Enumeration
• Excessive Agency: Unrestricted Tool Access, Autonomous Decision Making, Scope Creep
• Privilege Escalation: Excessive Permissions, Sudo/Root Execution, Credential Access
• Supply Chain: Unpinned Dependencies, External Script Fetching, Obfuscated Code

Findings (11)

Natural-Language Policy Violations

Medium

• Confidence: 94% confidence
• Finding: The file stores inferred or detected language data for specific contacts, including the source of that inference such as email, text, or WhatsApp history, without any indication of consent or necessity. Language inference is behavioral profiling and can be sensitive personal data in context; combined with names, relationship notes, religion-linked cultural context, and communication channel, it increases privacy risk and makes the contact dataset more intrusive if exposed or reused.

Missing User Warnings

Medium

• Confidence: 92% confidence
• Finding: The example explicitly references sensitive personal and social data sources such as conversation history, contact notes, calendar events, and shared LinkedIn/Instagram tabs without any prominent privacy notice, consent boundary, or data-minimization explanation. Even though the skill says it never sends messages automatically, aggregating and inferring from this data can expose relationship details, religious observance, birthdays, and communication patterns in ways users may not expect.

Natural-Language Policy Violations

Medium

• Confidence: 92% confidence
• Finding: This JSON assigns a specific output language based on prior communication history, such as "language": "en" with "language_source": "email_history". The file provides no indication that users can choose, override, or opt in to these locale decisions, which is a natural-language policy concern under the language/locale rule.

Natural-Language Policy Violations

Medium

• Confidence: 92% confidence
• Finding: This entry directs communication in Japanese based on "email_history" without any visible indication that the user selected or approved that language behavior. Because the file hard-codes locale behavior per contact and does not show an opt-in or override path, it fits the language/locale policy violation category.

Natural-Language Policy Violations

Medium

• Confidence: 92% confidence
• Finding: This record specifies Arabic based on WhatsApp history, but the file does not indicate user choice, opt-in, or a documented justification for enforcing that locale. Automatically inferring a communication language from prior history can violate language-choice policy when no override or consent is shown.

Natural-Language Policy Violations

Medium

• Confidence: 90% confidence
• Finding: This entry specifies French from text history without showing that the user explicitly chose or approved that language constraint. The locale decision appears automatic and fixed in the data, which is a policy issue unless accompanied by opt-in or clear documented justification.

Natural-Language Policy Violations

Medium

• Confidence: 95% confidence
• Finding: The template directs the system to generate outreach drafts in the contact's language based on inferred signals such as history, profile, or field, without requiring explicit user selection or consent. This can cause unintended impersonation, privacy, or cultural/relationship harms, especially if the inferred language is wrong or if the user did not intend multilingual outreach.

Natural-Language Policy Violations

Medium

• Confidence: 92% confidence
• Finding: The template explicitly instructs drafting in the language detected from the contact's communication history, which can override the user's preferred output language without an explicit opt-in. This can surprise users, expose incorrect inferences about language preference, and cause inappropriate or privacy-sensitive outreach if the detected language is wrong or contextually unsuitable.

Natural-Language Policy Violations

Medium

• Confidence: 96% confidence
• Finding: The template instructs the system to draft in the language detected from a contact's communication history, but it does not require explicit user confirmation or a fallback to the user's preferred output language. Inferring and acting on language preference from prior communications can create privacy and consent issues, and may lead to unintended outreach in a language the user did not want used in the current context.

Natural-Language Policy Violations

Medium

• Confidence: 91% confidence
• Finding: The template instructs the system to draft in the language detected from a contact's communication history, which relies on inferred personal attributes or preferences without explicit user confirmation. In a relationship/outreach tool, this can create privacy concerns, incorrect assumptions about identity or culture, and potentially sensitive profiling based on prior communications.

Scope Creep

Low

• Category: Excessive Agency

Content

## Multilingual

Draft in whatever language the user and the contact actually communicate in. This skill supports **any language** — it is not limited to a preset list.

**Language detection priority:**

• Confidence: 70% confidence
• Finding: Skill's behavior or capabilities extend beyond its stated purpose. Scope creep allows an agent to perform actions unrelated to its documented functionality, increasing the attack surface.

Static analysis

No suspicious patterns detected.