Back to skill

Security audit

Context Vault Manage

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a memory-vault guide, but it asks for broad command, write, network callback, and persistent memory powers without enough concrete limits.

Review before installing. This skill may be acceptable in a controlled local environment, but only if you are comfortable granting it persistent memory storage and file-write access. Avoid enabling unrestricted shell execution, external callbacks, or third-party embedding providers unless you can constrain commands, paths, destinations, credentials, retention, and deletion behavior yourself.

Vulnerability Patterns
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
  • Skill Instruction HijackingAlters the agent's session goals or safety constraints when the skill loads
  • Agent Memory PoisoningWrites attacker-controlled rules into memory that affect later sessions
  • 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 (1)

T05 · Unauthorized Access and Privilege Escalation

Warning
Location
SKILL.md:8
Finding

Unnecessary Command-Execution Capability Violates Least Privilege

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 8–13
Vulnerability Type: Excessive tool permissions
Risk Level: Medium

Vulnerable Code

yaml
tools:
  - read
  - exec
  - write
  - glob
  - grep

Technical Analysis

The Skill requests the unrestricted exec tool even though its documented memory-management workflows do not contain a concrete operation that requires shell execution. The declared functionality consists primarily of storing, searching, summarizing, and organizing memory. Granting command-execution capability therefore exceeds the minimum permissions demonstrated as necessary by the Skill documentation.

This broad capability increases the consequences of malicious input, unsafe future instructions, or operator error. If untrusted content processed by the Skill is interpreted as an instruction to invoke exec, commands could run with the operating-system privileges of the hosting Agent. The document claims that command execution is restricted by an allowlist and does not concatenate user input, but the audited project contains no implementation through which those controls can be verified.

Attack Path

  1. An operator loads the Skill, making its declared exec capability available.
  2. The Skill processes untrusted content, such as memory text, structured input, or instructions supplied by a user.
  3. Malicious or mistakenly interpreted content causes the Agent to invoke exec.
  4. Because no executable implementation or verifiable command allowlist is included, the effective command restrictions cannot be established from the audited package.
  5. A permitted command executes with the privileges and filesystem or network access of the Agent process.

This is a conditional exploitation path: the declaration alone does not execute commands, but it unnecessarily exposes a high-impact execution primitive.

Impact Assessment

Successful exploitation could allow command execution under the Agent process's acco ...[truncated 385 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove exec from the declared tool list unless a concrete, essential workflow requires operating-system command execution.
  2. Implement memory operations through narrowly scoped APIs rather than a general-purpose shell.
  3. If execution is indispensable, expose only fixed operations with explicit schemas and strict argument validation.
  4. Use an exact command-and-argument allowlist; reject shell metacharacters, command substitution, redirection, pipes, and unvalidated paths.
  5. Never construct shell commands by concatenating user-controlled memory content, queries, project names, callback URLs, or configuration values.
  6. Run permitted operations in a sandbox under a dedicated, unprivileged account with restricted filesystem and network access.
  7. Require explicit user confirmation before any security-sensitive command is executed.
  8. Add an auditable implementation and automated tests proving that untrusted input cannot alter executable commands.
  9. Replace generic advice to run with administrator privileges with least-privilege permission troubleshooting.
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Supply ChainUnpinned Dependencies, External Script Fetching, Obfuscated Code
Findings (6)

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The summary and description force Chinese-language interaction/documentation, and the rest of the file continues in Chinese. There is no indication that users may choose another language or that the skill is intentionally limited to a Chinese-speaking or region-specific context.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill is presented as a memory-vault manager, yet it introduces a callback URL parameter and declares powerful tools such as exec/write without clearly constraining when or why they are needed. This creates a capability/intent mismatch that can mislead users and downstream agents into granting broader execution or network behaviors than the documented core purpose requires.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The skill describes automatic summarization, cleanup, and archival of user memory without a clear warning that user data may be transformed, compressed, or moved automatically. In a memory-management skill, silent modification of stored context can affect integrity, privacy expectations, retention obligations, and downstream decision-making based on incomplete summaries.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The vector database and embedding sections describe generating embeddings and storing memory in LanceDB/Chroma or via external embedding models, but they do not clearly warn about privacy, data residency, provider exposure, or whether user content leaves the local environment. Because this skill manages potentially sensitive long-term memory, unclear embedding/storage handling materially raises confidentiality and compliance risk.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The troubleshooting and operational sections discuss generic command execution, permissions, timeouts, and network failures rather than narrowly scoped memory-management actions. In a skill that already requests exec, write, grep, and glob, this broad execution framing increases the chance that an agent or operator treats arbitrary shelling out as expected behavior.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

The security section claims protections such as HTTPS-only transport, SSL validation, and command whitelisting, but the file provides no enforceable mechanism or detailed policy showing those controls exist. Unsupported security assurances are dangerous because they can cause users or agents to trust network callbacks and command execution paths that may not actually be constrained.

Content

No source excerpt is available for this finding.