Back to skill

Security audit

idea-refine

Security checks for vulnerabilities and agentic risk

Overview

This skill appears to be a vague idea-refinement helper that asks for broad command, file, API, and credential-related authority without enough scoping.

Review before installing. Only use this skill in a tightly sandboxed workspace, do not grant shell execution or administrator privileges for idea refinement, and do not provide API keys or credentials unless the publisher supplies clear endpoints, scopes, and consent controls.

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 Command Execution Capability and Unsafe Privilege-Elevation Guidance

Content
View full analysis

Vulnerability Details

File Location: SKILL.md:17-19; related privilege-elevation guidance at SKILL.md:168
Vulnerability Type: Excessive tool permissions and privilege-escalation guidance
Risk Level: Medium

Vulnerable configuration and guidance:

yaml
tools:
  - read
  - exec
text
| Insufficient permissions | The current user lacks read/write permissions | Check file permissions and run as administrator |

Technical Analysis

The skill declares access to the general-purpose exec tool even though its stated purpose is refining ideas and processing textual input. No executable implementation or narrowly defined command requirement is present in the package. Consequently, arbitrary command execution is not demonstrated to be necessary for the advertised functionality and violates the principle of least privilege.

The troubleshooting section compounds this exposure by recommending execution with administrator privileges when permission errors occur. The recommendation does not identify a trusted command, limit the elevated operation, require user confirmation, or advise correcting narrowly scoped permissions first.

This does not prove that the skill itself automatically executes malicious commands. However, on an Agent platform that grants declared tools to loaded skills, the unnecessary capability creates a command-execution channel that could be influenced by malicious user input, prompt injection in processed content, or future unsafe instructions.

Attack Path

  1. A user installs or loads the skill on a platform that grants the tools declared in SKILL.md.
  2. The platform makes the unrestricted exec capability available during skill execution.
  3. An attacker supplies crafted content that attempts to induce the Agent to invoke shell commands.
  4. If the Agent follows the injected instruction without adequate confirmation or command restrictions, commands execute with the Ag ...[truncated 1103 chars]
Remediation
View remediation

Remediation Suggestions

  1. Remove exec from the declared tools because the documented idea-refinement workflow can operate through LLM text processing without shell access.
  2. If a future implementation genuinely requires command execution, replace unrestricted execution with a narrowly scoped tool or an explicit allowlist of fixed commands and validated arguments.
  3. Require clear user confirmation before every command that writes files, accesses sensitive locations, installs software, or changes system configuration.
  4. Run the Agent as a dedicated, unprivileged account with access restricted to the intended workspace.
  5. Replace the blanket recommendation to run as administrator with guidance to diagnose ownership and grant only the minimum required permission to the specific resource.
  6. Reject shell metacharacters and avoid constructing commands from untrusted content, format, or options values.
  7. Add platform-level controls for prompt-injection resistance, command logging, execution timeouts, filesystem isolation, and network restrictions.
Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (7)

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
98% confidence
Finding

The declared exec capability is not justified for refining ideas, which is normally a text-only task. Unnecessary command execution materially expands the attack surface: prompt-influenced content or downstream workflows could trigger local command execution, environment access, or chaining into more dangerous actions.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The summary and description combine Chinese and English phrasing, including mixed-language usage in the core skill description, without stating whether the user can choose their preferred language. This can create a de facto language policy or inconsistent locale behavior without explicit user opt-in.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The skill is presented as an idea-refinement tool, but its manifest broadens scope into generic development automation and orchestration. That mismatch can cause an agent or user to grant capabilities such as file access and execution under misleading pretenses, increasing the chance of unsafe invocation and privilege misuse.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The invocation guidance mixes several broad conditions such as using the skill when an idea is vague and when Development automation, data analysis, or process orchestration are needed. This scope is so expansive that it overlaps with many ordinary requests and does not clearly distinguish when the skill should or should not activate beyond one vague exclusion.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The title and primary label imply a focused idea-refinement skill, but the body describes broad automation and data-processing behaviors. This inconsistency is dangerous because trust and permission decisions may be made based on the benign name while the effective behavior is much more expansive.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The documentation claims support for GitHub-backed validation, external data ingestion, API connectivity, file handling, and command-oriented workflows that do not fit the stated idea-refinement use case. This kind of capability inflation can mislead operators and create covert paths for data access or exfiltration through systems that were never needed for the advertised function.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

Requiring API keys, credentials, and external data sources for an idea-refinement skill is unjustified and encourages overprivileged deployment. Even if no exploit is explicitly shown, collecting secrets and external connectivity without a strong functional need increases exposure to credential misuse, unintended data disclosure, and supply-chain risk.

Content

No source excerpt is available for this finding.