Security audit

engineering-drawing-prep

Security checks for vulnerabilities and agentic risk

Overview

This skill mostly fits its engineering document-prep purpose, but it has review-worthy mismatches around source-file mutation and overstated drawing standardization.

Review before installing. Use only on a copied client folder, keep original deliverables backed up or read-only, and do not treat the WIP DWGs as actually standardized until the script implements and verifies layer, font, title-block, and plot-setting changes.

SkillSpector

Vulnerability Patterns

MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
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
Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access

Findings (6)

Lp3

Medium

Category: MCP Least Privilege
Confidence: 89% confidence
Finding: The skill documentation instructs users to run local shell and Python scripts that read from a client folder and write standardized outputs, reports, tables, and RFIs, but the skill declares no permissions. This creates a transparency and policy-enforcement gap: an agent or platform may not properly warn users or constrain filesystem access, increasing the risk of unintended reading of sensitive client files or overwriting local data.

Description-Behavior Mismatch

Medium

Confidence: 96% confidence
Finding: The script modifies the client-provided source directory by deleting files during what is presented as an intake audit step. Even if the targeted files are considered 'noise', destructive behavior against original deliverables can cause loss of evidence, chain-of-custody issues, and unintended deletion of files the client expected to retain.

Intent-Code Divergence

Medium

Confidence: 92% confidence
Finding: The script header and stated purpose imply read-only auditing, but the implementation deletes files from the source tree. This mismatch increases operational risk because users may run it on original client deliverables expecting a safe inventory operation, leading to unexpected alteration of submitted materials.

Description-Behavior Mismatch

Medium

Confidence: 97% confidence
Finding: The script advertises DWG standardization but only copies files and emits a report, creating a dangerous mismatch between documented behavior and actual behavior. In an engineering workflow, downstream users may rely on the output as having been normalized for layers, fonts, or title blocks when it has not, which can propagate incorrect assumptions, QA failures, or unsafe design inputs.

Intent-Code Divergence

Medium

Confidence: 95% confidence
Finding: The inline comments state that DWG layers, fonts, and title blocks are standardized, but the code performs no such modifications. This kind of deceptive or stale documentation is risky in document-preparation pipelines because operators may skip manual review or trust outputs that were never brought into compliance.

Missing User Warnings

Medium

Confidence: 88% confidence
Finding: The README explicitly describes operations that strip files and standardize drawings, and the quick-start workflow runs those scripts directly without a clear warning that processing will create modified outputs and may alter copied deliverables if scripts are misused. In a document-preparation skill handling engineering project files, insufficient disclosure increases the risk of accidental data loss, overwrite, or users treating generated artifacts as authoritative without understanding the transformation step.

Static analysis

No suspicious patterns detected.