Back to skill

Security audit

Drawing Analyzer

Security checks for vulnerabilities and agentic risk

Overview

The skill appears intended for construction drawing analysis, but it asks for broad command execution and uses an unpinned external PDF dependency while its metadata under-discloses that setup.

Review this skill before installing. It is not showing exfiltration or destructive behavior, but only use it in an environment where you are comfortable letting the agent run Python commands and read construction PDFs. Prefer installing pdfplumber in an isolated virtual environment with pinned versions or a lock file, and treat extracted drawing metadata as potentially sensitive project information.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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)

T08 · Insecure Dependencies

Warning
Location
SKILL.md:481
Finding

Unpinned Third-Party Dependency Installation

Content
View full analysis

Vulnerability Details

File Location: SKILL.md, lines 481–483
Vulnerability Type: Unpinned third-party dependency installation
Risk Level: Medium

Vulnerable Code

bash
pip install pdfplumber

The installed dependency is subsequently imported and used at lines 60 and 151:

python
import pdfplumber
python
with pdfplumber.open(pdf_path) as pdf:

Technical Analysis

The setup instructions install pdfplumber without a fixed version or integrity hash. Consequently, installation resolves whichever compatible release and transitive dependencies are available from the configured Python package index at execution time. The reviewed Skill also declares the exec tool, enabling an Agent to run this installation command.

This creates a mutable software supply-chain boundary: future upstream changes, a compromised package index or publisher account, or dependency-resolution behavior involving an unsafe configured index could introduce unreviewed code. Python package installation can execute build backends and other installation-time components, while the resulting package executes during drawing analysis.

The audit found no evidence that pdfplumber is currently malicious, no typosquatted package name, and no intentionally unsafe package source. The risk arises from missing version and integrity controls rather than a confirmed malicious payload.

Attack Path

  1. A user or Agent follows the Skill's setup instructions.
  2. The declared exec capability runs pip install pdfplumber.
  3. Package resolution retrieves an unpinned release and its transitive dependencies from the environment's configured package index.
  4. If an upstream artifact, publisher account, index, or dependency-resolution source has been compromised, attacker-controlled package code may run during installation or later import.
  5. The installed dependency is imported and processes user-supplied drawing files under the Agent's operating-system identit ...[truncated 570 chars]
Remediation
View remediation

Remediation Suggestions

  1. Pin pdfplumber and every transitive dependency to versions that have been reviewed and tested.
  2. Maintain dependencies in a lock file generated through a reproducible dependency-management workflow.
  3. Require package hashes, such as with pip install --require-hashes -r requirements.txt, to verify artifact integrity.
  4. Install only from an explicitly trusted package index and disable unintended supplemental indexes.
  5. Perform installation in a dedicated, non-privileged virtual environment or isolated container.
  6. Scan locked dependencies for known vulnerabilities and review updates before changing pins.
  7. Separate setup from normal Skill execution and remove or restrict the exec capability after dependencies are provisioned.
  8. Process untrusted PDFs in a sandbox with restricted filesystem and network access to limit the effect of parser or dependency compromise.
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)

Vague Triggers

High
Category
Not specified by scanner
Confidence
95% confidence
Finding

The trigger keywords are very broad and overlap with common user requests like 'analyze', 'extract', and 'drawing', making accidental activation more likely. In a skill with read and exec capabilities, overbroad invocation increases the chance that the agent applies this skill in unintended contexts and exposes files or permissions unnecessarily.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

A substantial portion of the skill description is written in Chinese while the rest of the file is in English, and there is no indication that the user can choose the language or that the skill is intentionally limited to a Chinese-speaking context. This can violate language or locale policy when users are not given an explicit opt-in or locale justification.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The manifest makes strong safety claims such as removing risky code and external dependencies, but the skill still declares exec capability and documents use of the external pdfplumber library. This mismatch can mislead users or hosting platforms into granting trust or permissions under false assumptions, increasing the chance of unsafe deployment or review bypass.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

Declaring exec expands the skill's available attack surface beyond simple document parsing. Even though the sample code does not invoke shell commands, unnecessary command-execution privileges violate least-privilege principles and could be abused by future modifications, prompt-driven behavior, or host-agent integrations.

Content

No source excerpt is available for this finding.

Intent-Code Divergence

Low
Category
Not specified by scanner
Confidence
78% confidence
Finding

The dependency table says 'LLM API' is required, while the API key section states the skill is Markdown-based and requires no additional API key unless explicitly noted. These statements create an intent/documentation divergence about whether external API-backed capability is actually needed for this skill.

Content

No source excerpt is available for this finding.