Back to skill

Security audit

context-engineering

Security checks for vulnerabilities and agentic risk

Overview

This skill appears to be a broad automation template with command-execution authority, despite being presented as a context-engineering helper.

Review this skill carefully before installing. It does not contain an executable payload, but it asks for command execution and describes broad automation, credentials, files, and network/API use without narrow limits. Only install it in an environment where command execution is sandboxed or disabled, and do not provide secrets or elevated permissions unless the skill is rewritten with clear scope and confirmation rules.

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:17
Finding

Unnecessary Unrestricted Command-Execution Capability

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 17-19
Vulnerability Type: T05: Unauthorized Access and Privilege Escalation
Risk Level: Medium

Vulnerable Code:

yaml
tools:
  - read
  - exec

Technical Analysis

The Skill requests the generic exec tool despite documenting only context configuration, content parsing, data transformation, and structured result generation. The package provides no executable implementation, command allowlist, argument validation rules, shell-interpolation restrictions, sandbox boundary, or user-confirmation requirement that would justify or constrain this capability.

This violates the principle of least privilege. If the hosting Agent grants the declared tool permissions, the Skill receives a command-execution channel beyond the needs demonstrated by its documented functionality. The risk is amplified by the Skill's broad activation conditions and its acceptance of user-controlled content.

Attack Path

  1. The Skill is installed on an Agent that grants tools declared in SKILL.md.
  2. A user or attacker submits crafted content that falls within the Skill's broadly described processing scenarios.
  3. The Skill is activated with access to the unrestricted exec tool.
  4. Because no command allowlist, argument validation, or confirmation boundary is defined, crafted instructions may be interpreted as a request to invoke local commands.
  5. Any resulting command executes with the operating-system permissions and environment access of the Agent process.

No direct malicious command or executable payload is present in the audited package; exploitation is conditional on the hosting Agent exposing exec and allowing untrusted content to influence tool use.

Impact Assessment

Successful exploitation could allow commands to run with the privileges of the Agent process. Depending on the host configuration, this may expose readable local files, environme ...[truncated 419 chars]

Remediation
View remediation

Remediation Suggestions

  1. Remove exec from the declared tools and retain only capabilities required for context and text processing.
  2. If command execution is essential, replace unrestricted execution with a narrowly scoped operation or explicit command allowlist.
  3. Pass validated arguments directly to approved executables without invoking a shell or constructing command strings through interpolation.
  4. Reject metacharacters, unexpected options, untrusted paths, and arguments outside documented schemas.
  5. Run approved commands in a least-privilege sandbox with restricted filesystem, environment-variable, process, and network access.
  6. Require explicit user confirmation before any command that reads sensitive data, modifies files, accesses the network, or changes system state.
  7. Document every permitted command, its purpose, accepted arguments, side effects, timeout, and failure behavior.
  8. Add security tests verifying that user-controlled content cannot select arbitrary executables or inject additional command arguments.
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 (6)

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The manifest presents the skill as a narrow context-setup utility, but the body describes much broader automation capabilities including API connectivity, command execution, and file handling. This mismatch can mislead the agent or user into granting trust, tools, or invocation in situations where the real behavior is materially more powerful, increasing the risk of over-privileged or unexpected actions.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The core-function section contradicts the advertised purpose by describing a generic data-processing and automation skill rather than context engineering. In a skill ecosystem, semantic drift like this is dangerous because routing, policy review, and user expectations may be based on the declared purpose while the actual instructions expand into unrelated, potentially riskier operations.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The quick-start and troubleshooting sections introduce API keys, network access, file operations, and command execution, which are not consistent with a narrowly scoped context-engineering skill. This can normalize sensitive operations without adequate warning, causing users or agents to perform state-changing or secret-bearing actions under a misleading label.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The documentation references credential setup, network/API use, file processing, and command execution without prominently explaining security implications such as secret exposure, external data transfer, or local system modification. In practice, this can cause users or agents to undertake sensitive actions without informed consent or sufficient safeguards.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The trigger condition is so broad that the skill may appear eligible for nearly any related request, increasing the chance of accidental or overbroad activation. For a skill with declared exec capability and documentation mentioning files, commands, and external resources, vague activation criteria materially raise misuse risk.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

The summary and description include Chinese text and mixed-language instructions, but the document does not state that the skill is China-specific or let users choose their preferred language. This can violate language/locale policy when a skill implicitly forces a language experience without opt-in.

Content

No source excerpt is available for this finding.