Back to skill

Security audit

Simple Memory Skill

Security checks for vulnerabilities and agentic risk

Overview

The skill is a local memory helper, but it under-discloses a global third-party npm install and makes broad, persistent memory logging the default.

Review before installing. This skill is not clearly malicious, but it asks your agent to install an unpinned global npm package and to persist conversation details across sessions. Use it only in a workspace where local memory files are acceptable, avoid storing secrets or sensitive personal data, and prefer a pinned, reviewed, project-local CLI with explicit opt-in memory writes.

Vulnerability Patterns
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Remote Payload Retrieval and ExecutionFetches external code whose behavior can change after review
  • Embedded Malicious CodeShips malicious scripts inside the skill and executes them locally
Findings (2)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:128
Finding

Unpinned Third-Party Package Installed Globally

Content
View full analysis
Remediation
View remediation
``` 2. Publish and verify the expected package integrity hash or provenance before installation. 3. Link to the authoritative package source, publisher identity, and source commit corresponding to the pinned release. 4. Prefer bundling the required implementation in the Skill so that reviewers can inspect the code executed by the documented commands. 5. Review package lifecycle scripts and use `--ignore-scripts` if the package does not legitimately require installation scripts. 6. Avoid global installation where possible. Use a locked, project-local dependency with a committed lockfile and invoke the local executable. 7. Add automated dependency scanning and periodically review package ownership, release provenance, and advisories. 8. Correct the “zero-dependency” documentation so users understand that setup retrieves and executes a third-party package. ]]>

T02 · Agent Memory Poisoning

Error
Location
SKILL.md:146
Finding

Untrusted User Content Persisted and Reloaded as Agent Memory

Content
View full analysis
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (8)

Vague Triggers

High
Category
Not specified by scanner
Confidence
96% confidence
Finding

The trigger keywords include very generic terms like 'memory', 'simple', 'system', 'zero', and 'keys', which are likely to match ordinary conversation. Overbroad triggers can cause accidental activation of a skill with read/exec capabilities, increasing the chance of unintended file writes or command execution.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The manifest markets the skill as zero-dependency and pure local storage, yet the documented workflow depends on installing and invoking a third-party CLI. This is dangerous because it obscures the real trust boundary and may cause users or automated systems to approve execution under false assumptions.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
81% confidence
Finding

The file is primarily in English but includes important descriptive and trigger-related content in Chinese, without indicating that users may choose their preferred language. This can amount to an implicit language policy choice because some users may be forced to interpret critical skill details in a language they did not opt into.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill claims risk code and external dependencies were removed, but the setup still requires globally installing and executing an external npm package. That mismatch can mislead users and agents into granting execution trust they would not otherwise give, increasing supply-chain and command-execution risk.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The examples direct creation and modification of local files such as SESSION-STATE.json, MEMORY.md, and the memories directory without prominently warning that the skill changes the filesystem. Users may unintentionally permit persistent file writes in sensitive directories or repositories.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill instructs the agent to persist user information locally as part of normal operation without clear consent, retention, or sensitivity boundaries. This creates a privacy and data-handling risk because personal preferences, decisions, and contextual details may be stored automatically and indefinitely.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill explicitly instructs the agent to save user-provided information before responding, which amounts to default logging and retention of conversational data. In the context of an AI agent with read/exec tools, this is more dangerous because it operationalizes systematic storage without requiring case-by-case approval.

Content

No source excerpt is available for this finding.

Ssd 3

Medium
Category
Not specified by scanner
Confidence
98% confidence
Finding

The WAL protocol mandates storing multiple categories of user input—preferences, decisions, deadlines, and corrections—before any reply, creating comprehensive interaction logs. This is dangerous because it normalizes broad data capture and persistence, potentially collecting sensitive information and making unintended disclosure more likely.

Content

No source excerpt is available for this finding.