Security audit

Huimemory Integration

Security checks for vulnerabilities and agentic risk

Overview

This is a coherent HuiMemory integration guide, but it needs Review because it handles long-term conversation history and examples can send retrieved memories to an external LLM without enough privacy guardrails.

Install or adapt this only if you intentionally want HuiMemory to index and retrieve historical conversations. Add explicit user consent before recall, restrict memory by user/session, define deletion and retention rules, redact sensitive content, and avoid sending recalled history to hosted LLM APIs unless users have clearly accepted that data transfer.

SkillSpector

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (5)

Description-Behavior Mismatch

Medium

Confidence: 95% confidence
Finding: The documentation frames HuiMemory as a local memory integration, but the sample initializes a remote OpenAI client and is positioned to send retrieved memory-derived content into an external model workflow. This creates a real risk of unintentional data exfiltration because developers may assume the integration remains fully local when conversation history can leave the device.

Description-Behavior Mismatch

Medium

Confidence: 98% confidence
Finding: The RAG example directly injects formatted historical dialogue into the prompt sent to an external chat completion API. If users store sensitive personal, business, or regulated data in memory, this pattern can leak that data to a third-party provider without clear consent or warning.

Vague Triggers

Medium

Confidence: 92% confidence
Finding: The activation guidance is very broad and instructs the agent to proactively use the skill whenever common memory- or retrieval-related keywords appear. That can cause unintended invocation in contexts where the user did not ask to access stored conversation history, increasing the risk of over-collection, irrelevant retrieval, or disclosure of prior sensitive exchanges.

Missing User Warnings

Medium

Confidence: 95% confidence
Finding: The skill promotes storing and retrieving historical conversations and long-term memory but does not warn about privacy, retention, consent, or exposure of sensitive data. In a memory system context, this omission is more dangerous because the entire feature centers on preserving and resurfacing user conversations, which may include personal, confidential, or regulated information.

Missing User Warnings

Medium

Confidence: 97% confidence
Finding: The examples normalize sending historical conversation memory to an external LLM but omit an explicit discussion of privacy, data transfer, and retention implications. In a skill marketed around local semantic memory, that omission materially increases the chance that integrators will expose sensitive history beyond the local environment.

Static analysis

No suspicious patterns detected.