Security audit

Memscape

Security checks for vulnerabilities and agentic risk

Overview

Memscape is a coherent memory-sharing skill, but it needs review because it encourages storing and sometimes publishing user and project context without clear safeguards.

Install only if you are comfortable with selected agent context being stored by Memscape and possibly shared publicly. Require the agent to ask before saving or promoting anything, review the exact text first, and avoid secrets, credentials, customer data, personal details, internal URLs, unreleased code details, and confidential project context.

SkillSpector

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep

Findings (6)

Missing User Warnings

Medium

Confidence: 97% confidence
Finding: The skill explicitly encourages agents to persist decisions, preferences, pitfalls, and handoffs across sessions, but it does not provide clear guardrails against storing secrets, personal data, regulated data, or confidential project information. Because the service is designed for cross-session retention and some data may later be shared or promoted, this creates a meaningful risk of privacy leakage and unauthorized retention of sensitive user-derived information.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The sharing guidance and examples focus on what is useful to contribute but omit warnings to avoid user-specific, confidential, or sensitive information. In a collective-knowledge system, this omission is dangerous because agents may generalize from the examples and upload proprietary context, internal troubleshooting details, or personal preferences tied to identifiable users.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The documentation encourages storing private memories, handoffs, preferences, and project context on a remote service but does not warn users against sending secrets, regulated data, proprietary code, or sensitive project details. In an agent skill context, this materially increases the risk of unintentional exfiltration of confidential information across session boundaries to a third-party service.

Ssd 3

Medium

Confidence: 98% confidence
Finding: The examples include storing and sharing user-specific preferences and conversational interpretations such as what a human 'actually means,' which encourages cross-session profiling of users. That is risky because these details can be sensitive, inaccurate, or context-dependent, and retaining or sharing them without consent can violate privacy expectations and influence future interactions in opaque ways.

Ssd 3

Medium

Confidence: 98% confidence
Finding: The top-level description broadly instructs the agent to retain preferences, decisions, patterns, pitfalls, and handoffs across sessions without defining retention limits, sensitivity classes, or approval requirements. In context, this is more dangerous because the core purpose of the skill is persistent memory, so the absence of privacy boundaries is central rather than incidental.

Ssd 3

Medium

Confidence: 97% confidence
Finding: The private memory, recall, handoff, resume, and promote flows all encourage retaining user preferences and session-derived context, yet they do not impose safeguards such as minimization, consent, redaction, or scope restrictions. This creates a substantial chance that sensitive details are retained longer than necessary and later surfaced to future sessions or promoted beyond the original context.

Static analysis

No suspicious patterns detected.