Back to skill

Security audit

writing-plans

Security checks for vulnerabilities and agentic risk

Overview

This skill is not clearly malicious, but it asks for command execution while describing a broad and inconsistent mix of planning, automation, data processing, and API workflows.

Review this skill carefully before installing. It does not show exfiltration, destructive behavior, hidden persistence, or bundled executable code, but its scope is too broad for a planning skill and it requests shell execution without clear limits. Prefer a version that removes exec or defines exactly which commands and workflows are allowed.

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 (5)

Description-Behavior Mismatch

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The manifest presents this as a pre-code planning skill, but the body expands it into generic automation, data processing, visualization, API use, file handling, and command execution. This scope mismatch can cause the agent to invoke the skill in contexts far beyond user expectations, enabling unnecessary access to sensitive data or operational capabilities.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

High
Category
Not specified by scanner
Confidence
99% confidence
Finding

Granting exec to a writing/planning skill violates least privilege because planning tasks do not require shell or process execution. If the agent routes tasks here based on the broad description, this unnecessary capability could be abused to run commands, touch files, or interact with the environment without a legitimate planning need.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The documentation advertises planning but the implementation narrative describes broad operational workflows such as automation, data handling, and command-related behavior. This inconsistency increases the chance of unsafe invocation and user deception, because reviewers and routing logic may trust the planning label while the content normalizes more powerful actions.

Content

No source excerpt is available for this finding.

Vague Triggers

High
Category
Not specified by scanner
Confidence
92% confidence
Finding

The invocation criteria are overly broad and ambiguous, covering project management, development automation, data analysis, reporting, and visualization. Broad triggers increase the likelihood that the agent will select this skill in inappropriate contexts, especially dangerous here because the skill also declares exec and reads as a catch-all operational tool.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
83% confidence
Finding

The description states Chinese interaction support in a way that may bias routing or output language without explicit user preference. While not a direct code-execution risk, forced or assumed language behavior can mis-handle user intent, reduce transparency, and make review of risky actions harder in multilingual environments.

Content

No source excerpt is available for this finding.