Security audit

本地长记忆

Security checks for vulnerabilities and agentic risk

Overview

This is a local memory skill, but it needs review because it directs agents to persist potentially sensitive user details without enough consent, retention, or deletion safeguards.

Install only if you want the agent to keep long-term local memory. Before using it with medical, financial, legal, business, or personal data, set clear rules for what must never be stored, protect the memory directory, and back up before running forget, cleanup, compact, or dedup commands.

SkillSpector

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 (5)

Missing User Warnings

Medium

Confidence: 87% confidence
Finding: The skill explicitly encourages storing highly sensitive data such as medical allergy information in persistent local memory, but does not require explicit user consent, retention limits, encryption, access controls, or privacy warnings. In privacy-sensitive environments, this can lead to silent long-term retention of regulated or confidential information, increasing the risk of local disclosure, unauthorized access, or compliance violations.

Missing User Warnings

Medium

Confidence: 84% confidence
Finding: The documented forget/delete command is destructive and presented without confirmation, dry-run behavior, or warning about irreversible loss. This creates a realistic risk of accidental or overbroad deletion of memory records, especially when an agent or user issues ambiguous queries.

Missing User Warnings

Medium

Confidence: 89% confidence
Finding: Cleanup, compaction, deduplication, cache cleaning, and backup-related maintenance commands can alter or remove stored data, yet the skill gives no strong warnings about irreversibility, selection criteria, or pre-backup requirements. In an agent-driven context, this raises the chance of unintended data loss or silent destruction of memory artifacts.

Ssd 3

Medium

Confidence: 90% confidence
Finding: The workflow instructs the agent to persist user-provided details, decisions, deadlines, and corrections across multiple local files and memory stores by default. Without data minimization, classification, consent, or sensitivity filtering, this creates a substantial risk of retaining secrets, personal data, or confidential business information beyond user expectations, making later disclosure or misuse more likely.

Ssd 3

Medium

Confidence: 90% confidence
Finding: The conversation lifecycle directs the agent to write concrete user details and session summaries into persistent artifacts such as SESSION-STATE.md, MEMORY.md, and daily logs. In context, this is more dangerous because the skill markets itself for privacy-sensitive domains, so operators may trust it with regulated data even though the instructions do not enforce minimization, consent, or secure handling controls.

Static analysis

No suspicious patterns detected.