Back to skill

Security audit

md-to-medium

Security checks for vulnerabilities and agentic risk

Overview

The scanners classify this skill as suspicious because its image-readiness claims do not match reported behavior and Chromium renders Markdown tables without its sandbox; review coverage is partial.

Consider the suspicious verdict and the reported browser exposure and image-check gaps before installing. Skillspector scanned both listed components, but its analysis is marked partial because several referenced paths were missing, so the evidence is incomplete.

Vulnerability Patterns
  • 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
  • Unauthorized Access and Privilege EscalationObtains permissions beyond the task's legitimate needs
Vulnerability Patterns
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (9)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The primary conversion purpose matches, but the declared image-safety guarantees do not. In convert(), a data: image only adds a note and is left unchanged; it is not added to the images list checked by main(), so it can produce a successful payload containing precisely the image source the description says is prohibited. Existing HTTP(S) images also bypass hosting checks. For local images, hosting is inferred from a locally available Git ref rather than verified against the public URL, and --allow-unpushed explicitly permits success despite blocked images. The script also writes the payload before enforcing its blocking exit status. These are substantive limitations of the advertised readiness guarantee, not evidence of an unrelated or malicious primary purpose.

Content

No source excerpt is available for this finding.

Anti-Refusal Statement

High
Category
Anti-Refusal
Confidence
80% confidence
Finding

Skill instructs the agent to omit warnings, disclaimers, or ethical commentary. Stripping safety caveats hides risk from the user and is a common jailbreak preamble.

Content

Scanner excerpt · SKILL.md (reported line 124)May include surrounding context.

md
A payload with `data:image` in it is the wrong payload. Then, after pasting, count the images in the draft — the editor gives no warning when one goes missing.

Skill declares no tool scope ('permissions' or 'allowed-tools') but code capabilities were detected: file_read, file_write, shell.

Medium
Category
MCP Least Privilege
Confidence
70% confidence
Finding

Without declared permissions the skill's intent is opaque and cannot be validated.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/md2medium.py (reported line 120)May include surrounding context.

python
hosted. Pixels are the honest comparison.
    """
    try:
        done = subprocess.run(
            ["magick", "compare", "-metric", "AE", str(a), str(b), "null:"],
            stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL, timeout=60,
        )

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
98% confidence
Finding

Table HTML from the Markdown is opened in Chromium before attributes and unsupported elements are sanitized, so malicious event handlers or embedded content can execute or initiate network requests during conversion. Chromium is also launched with --no-sandbox, removing an important containment boundary and increasing the consequences of a browser exploit against an untrusted draft.

Content

Scanner excerpt · scripts/md2medium.py (reported line 147)May include surrounding context.

python
new_path = out.with_name(f".{out.stem}.new.png")
    html_path.write_text(_table_page(str(table), out.stem), encoding="utf-8")

    subprocess.run(
        ["chromium", "--headless", "--no-sandbox", "--disable-gpu",
         "--hide-scrollbars", "--force-device-scale-factor=2",
         f"--window-size={width},4000", f"--screenshot={png_path}",

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/md2medium.py (reported line 155)May include surrounding context.

python
stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL,
        timeout=120, check=True,
    )
    subprocess.run(
        ["magick", str(png_path), "-trim", "+repage",
         "-bordercolor", "white", "-border", "12", str(new_path)],
        stdout=subprocess.DEVNULL, stderr=subprocess.DEVNULL,

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/md2medium.py (reported line 170)May include surrounding context.

python
def _git(repo: Path, *args: str) -> subprocess.CompletedProcess:
    return subprocess.run(["git", "-C", str(repo), *args],
                          capture_output=True, text=True)

Data-URI images remain in the output and bypass the promised readiness check.

Medium
Category
Excessive Agency
Confidence
99% confidence
Finding

The manifest promises that every image points to a public URL and that the converter refuses to pass until images are hosted. At L293–L295, a data-URI image only generates a note: its source remains unchanged and it is excluded from the images checked at L379–L396. Consequently, a document containing such an image can exit successfully and be copied to the clipboard even though Medium will silently delete the image.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
70% confidence
Finding

subprocess module calls execute external commands. Without careful input validation, this enables command injection.

Content

Scanner excerpt · scripts/md2medium.py (reported line 439)May include surrounding context.

python
if args.clip and not args.print_html:
        try:
            proc = subprocess.Popen(
                ["wl-copy", "-f", "--type", "text/html"],
                stdin=subprocess.PIPE, stdout=subprocess.DEVNULL,
                stderr=subprocess.DEVNULL, text=True, start_new_session=True,