Security audit

Long-Term Memory Manager

Security checks across malware telemetry and agentic risk

Overview

This is a useful local memory tool, but it encourages persistent retention and reuse of sensitive conversation, account, and credential-related information without enough user controls.

Install only if you intentionally want persistent local memory. Do not store passwords, tokens, OAuth links, financial details, or unnecessary personal data in these files; review MEMORY.md and daily logs before consolidation, archiving, summary export, or vector sync, and keep a backup before running mutating commands.

SkillSpector

By NVIDIA

Vulnerability Patterns

Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
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

Findings (7)

Context-Inappropriate Capability

Medium

Confidence: 90% confidence
Finding: `generate_summary()` accepts an arbitrary `output_file` path and writes to it directly via `Path(output_file).write_text(...)` with no restriction that the destination remain inside the memory workspace. In a skill intended for memory maintenance, this broad file-write capability can overwrite user files outside the managed area if an untrusted caller supplies a path.

Missing User Warnings

Medium

Confidence: 97% confidence
Finding: The skill explicitly recommends storing sensitive long-term facts such as account details and credential locations in MEMORY.md without privacy safeguards, minimization guidance, or consent requirements. Persistent retention of this kind of information increases the blast radius of any compromise, accidental disclosure, overbroad search, or unintended model reuse.

Missing User Warnings

Medium

Confidence: 93% confidence
Finding: The archival workflow modifies and moves memory data automatically, but the skill does not clearly warn that files will be changed, relocated, or propagated to other stores. This is dangerous because users may unknowingly trigger destructive or privacy-impacting operations that alter retention, visibility, and recoverability of personal information.

Ssd 3

Medium

Confidence: 96% confidence
Finding: The skill encourages durable storage of sensitive personal and operational information in a long-term memory file, including examples that normalize retention of account and credential-related metadata. In context, this is more dangerous because the skill's entire purpose is persistence, searchability, and consolidation, which amplifies exposure and secondary use risks over time.

Ssd 3

Medium

Confidence: 91% confidence
Finding: The daily memory guidance instructs logging important conversations and extracting user facts into persistent storage, which can capture personal data beyond what is necessary for task completion. Even if intended for convenience, this creates privacy and profiling risks because conversational content may be retained indefinitely and later searched or repurposed.

Ssd 4

Medium

Confidence: 95% confidence
Finding: The workflow describes moving conversational content from daily logs into MEMORY.md and then syncing key facts into vector memory, creating multiple persistent and searchable copies. This materially increases privacy risk because sensitive content can spread across systems with different access patterns, making deletion, auditing, and least-privilege enforcement much harder.

Ssd 3

Medium

Confidence: 93% confidence
Finding: The extraction logic explicitly treats `账号|account|credential` as important facts and then persists matched content into `MEMORY.md`, while `search()` can later return those values with surrounding context. In a long-term memory tool, this materially increases exposure and retention of secrets or account data, expanding the blast radius of accidental disclosure, compromise, or over-broad recall.

VirusTotal

66/66 vendors flagged this skill as clean.

View on VirusTotal

Static analysis

No suspicious patterns detected.