Security Audit

screenshot-feature-extractor

github.com/davila7/claude-code-templates

AI SkillCommit 458b11867eae

TRUSTED

Scanned 2 days ago

Critical

Immediate action required

High

Priority fixes suggested

Medium

Best practices review

Low

Acknowledged / Tracked

Trust Assessment

screenshot-feature-extractor received a trust score of 76/100, placing it in the Mostly Trusted category. This skill has passed most security checks with only minor considerations noted.

SkillShield's automated analysis identified 3 findings: 0 critical, 1 high, 1 medium, and 1 low severity. Key findings include Network egress to untrusted endpoints, Covert behavior / concealment directives, Potential Data Exfiltration via Arbitrary File Path in Agent Prompts.

The analysis covered 4 layers: Manifest Analysis, Static Code Analysis, Dependency Graph, LLM Behavioral Safety. All layers scored 70 or above, reflecting consistent security practices.

Last analyzed on February 20, 2026 (commit 458b1186). SkillShield performs automated 4-layer security analysis on AI skills and MCP servers.

Layer Breakdown

Manifest Analysis

91%

Static Code Analysis

100%

Dependency Graph

100%

LLM Behavioral Safety

85%

Behavioral Risk Signals

Network Access

1 finding

Filesystem Write

1 finding

Shell Execution

1 finding

Security Findings3

Severity	Finding	Layer	Location
HIGH	Potential Data Exfiltration via Arbitrary File Path in Agent Prompts The skill constructs prompts for sub-agents (`screenshot-ui-analyzer`, `screenshot-interaction-analyzer`, `screenshot-business-analyzer`) by directly embedding a user-provided `[file path]` into the prompt (e.g., `Screenshot: [file path]`). If a malicious user provides a path to a sensitive system file (e.g., `/etc/passwd`, `~/.aws/credentials`), and the underlying agent execution environment or the LLM itself has file reading capabilities, the content of that file could be read and included in the agent's JSON analysis output, leading to data exfiltration. The prompt explicitly instructs the agent to 'Analyze this screenshot' and then provides the path, which an LLM might interpret as an instruction to read and process the content of the specified file. Implement strict input validation and sanitization for file paths. Ensure that file paths provided by users are restricted to a designated, sandboxed directory. Instead of passing the raw file path directly to the LLM, pass a file ID or a pre-loaded content representation. The agent should only receive the content of the image, not a path it can interpret for arbitrary file system access. Ensure the underlying file access mechanism is sandboxed and only allows access to explicitly uploaded and validated files.	LLM	SKILL.md:58
MEDIUM	Network egress to untrusted endpoints HTTP request to raw IP address Review all outbound network calls. Remove connections to webhook collectors, paste sites, and raw IP addresses. Legitimate API calls should use well-known service domains.	Manifest	cli-tool/components/mcps/devtools/figma-dev-mode.json:4
LOW	Covert behavior / concealment directives Multiple zero-width characters (stealth text) Remove hidden instructions, zero-width characters, and bidirectional overrides. Skill instructions should be fully visible and transparent to users.	Manifest	cli-tool/components/mcps/devtools/jfrog.json:4

Scan History

Embed Code

[![SkillShield](https://skillshield.io/api/v1/badge/c20fff397b61cd39.svg)](https://skillshield.io/report/c20fff397b61cd39)