Back to skill

Security audit

mobile-reading-export

Security checks for vulnerabilities and agentic risk

Overview

ClawScan marks this skill suspicious because scanners found that its Chrome rendering disables a browser safety boundary and that PDF replacement may accept an invalid output; analysis was partial.

Consider the suspicious verdict and the confirmed rendering and PDF replacement weaknesses before installing. Skillspector inspected all eight listed components, but reported partial analysis because three local references were unresolved; the evidence is therefore 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 Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (4)

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 119)May include surrounding context.

md
### Windows + headless Chrome (the ones that cost real time)

- **`--window-size` has a minimum on Windows (~552 CSS px).** Asking for `393` silently yields a ~526px layout viewport while the screenshot is still 393px wide — content is clipped on *both* sides (the left side loses the start of each wrapped line). Chrome emits no error and no warning for this clamp, so you must actively verify the rendered width (step 3) and warn the user that a silently clipped layout is possible.
- **`--force-device-scale-factor` scales the PNG but does not shrink the CSS viewport.** `--window-size=393,850 --force-device-scale-factor=2` gives a 786×1700 PNG of a 526px-wide layout.
- **Without `--window-size`, Chrome's default layout viewport is ~754px.** Any measurement taken without passing the window size is measuring the wrong viewport.
- **A full-width coloured probe cannot detect the clamp.** A bar spanning the full viewport, clipped to the image width, measures exactly the image width. Measure a

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
97% confidence
Finding

The subprocess uses a fixed executable and argument list, so the invocation does not establish shell injection. However, its callers explicitly pass --no-sandbox while rendering supplied HTML, disabling an important containment boundary for potentially hostile document content. A browser vulnerability could consequently compromise the rendering account with fewer isolation barriers.

Content

Scanner excerpt · scripts/cap.py (reported line 69)May include surrounding context.

python
def chrome(args):
    r = subprocess.run([CHROME] + args, capture_output=True)
    return r.stdout.decode('utf-8', 'replace'), r.stderr.decode('utf-8', 'replace'), r.returncode

The overwrite message promises to preserve the original until a valid PDF is verified, but replacement requires only a nonempty output file.

Medium
Category
Excessive Agency
Confidence
98% confidence
Finding

Despite promising to retain the original until a valid PDF is verified, the code replaces it whenever the staged file is nonempty and the subsequent parsing does not raise an exception. It neither checks Chrome's exit status nor requires valid PDF structure or any pages, so corrupt or incomplete output can destroy an existing deliverable when overwrite is authorized.

Content

No source excerpt is available for this finding.

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
97% confidence
Finding

The fixed Chrome executable and list-form arguments do not introduce shell injection, but the command explicitly disables Chrome's sandbox when loading supplied HTML. In this document-conversion context, attacker-controlled HTML can reach the browser engine without that containment boundary, increasing the consequences of a browser exploit.

Content

Scanner excerpt · scripts/topdf.py (reported line 25)May include surrounding context.

python
fd, staged = tempfile.mkstemp(suffix='.pdf', prefix='.', dir=os.path.dirname(os.path.abspath(out)))
    os.close(fd)
    url = 'file:///' + os.path.abspath(src).replace('\\', '/')
    r = subprocess.run([CHROME, '--headless=new', '--disable-gpu', '--no-sandbox',
                        '--no-pdf-header-footer', '--virtual-time-budget=6000',
                        f'--print-to-pdf={staged}', url],
                       capture_output=True)