Back to skill

Security audit

RSS转MD(专业版)

Security checks for vulnerabilities and agentic risk

Overview

The skill is a disclosed RSS-to-Markdown archiving assistant, but it asks for broad execution authority and includes under-scoped scheduled, file-writing, and document-conversion examples that deserve user review.

Install only if you are comfortable with an agent fetching RSS URLs, writing archive/export files, optionally using an LLM, and running local conversion tools. Use trusted feeds, keep outputs in a dedicated folder, avoid enabling scheduled archiving until you have reviewed the paths, and sandbox or skip PDF/EPUB conversion for untrusted content.

Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (7)

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
84% confidence
Finding

The skill advertises file saving, scheduled archiving, image downloading, and full-content retrieval while exposing powerful tools including exec, yet it does not consistently require explicit user confirmation, destination/path constraints, or safe defaults for data-impacting actions. In an agent context, that can lead to unintended persistent writes, repeated scheduled activity, or broad remote content retrieval with privacy, storage, and operational consequences.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
87% confidence
Finding

The examples write files to user-supplied paths and perform multi-format exports without any warning, consent flow, or path restrictions. In an agent skill context, silent local file modification is dangerous because users may trigger writes unintentionally, and downstream components may overwrite sensitive files or create artifacts in unexpected locations.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
90% confidence
Finding

The batch conversion example fetches arbitrary feed URLs over the network without a user-facing warning or apparent allowlisting. In an agent environment, silent outbound requests can leak metadata, contact attacker-controlled hosts, or be abused for SSRF-like access if internal URLs are accepted.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
94% confidence
Finding

The examples invoke external utilities for PDF/EPUB conversion without warning the user that local executables will be run. In a skill context, unexpected subprocess execution increases attack surface, can process attacker-controlled content with complex parsers, and may expose the host to vulnerabilities in tools like wkhtmltopdf or pandoc.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
83% confidence
Finding

The AI enhancement example sends feed content into an LLM-like processing path without warning users about possible data disclosure. Even though the current implementation is stubbed, the design pattern normalizes forwarding potentially sensitive content to an AI component, which becomes risky once connected to a real external model service.

Content

No source excerpt is available for this finding.

Unvalidated Output Injection

High
Category
Output Handling
Confidence
88% confidence
Finding

The code passes generated HTML derived from feed content into wkhtmltopdf, a powerful external renderer with a history of risky behaviors around active content, local file access, and network retrieval depending on configuration. Because the HTML embeds unescaped feed title and markdown content directly, an attacker controlling feed data may be able to trigger unsafe renderer behavior, data exposure, or processing of hostile content when PDF export is run.

Content

Scanner excerpt · references/detail.md (reported line 237)May include surrounding context.

md
import subprocess
        try:
            subprocess.run(['wkhtmltopdf', html_path, output_path], check=True)
            return output_path
        except Exception as e:
            print(f"PDF转换失败:{e}")

Unvalidated Output Injection

High
Category
Output Handling
Confidence
81% confidence
Finding

The code invokes pandoc on markdown generated from untrusted feed content and writes to a caller-controlled output path. While there is no shell injection here, feeding attacker-controlled content into a complex document conversion tool can expose parser/renderer attack surface and may permit unintended file access or unsafe includes depending on pandoc features and environment.

Content

Scanner excerpt · references/detail.md (reported line 250)May include surrounding context.

md
import subprocess
        try:
            subprocess.run(['pandoc', md_path, '-o', output_path], check=True)
            return output_path
        except Exception as e:
            print(f"EPUB转换失败:{e}")