Back to skill

Security audit

Converter

Security checks for vulnerabilities and agentic risk

Overview

The verdict is benign because the reported findings are assessed as consistent with a file-conversion skill, though the scan has medium confidence and incomplete reference coverage.

The verdict is benign, not a guarantee of zero risk. The scan covered all four listed components, but it was marked partial because two local path-like references in SKILL.md could not be resolved, so evidence is incomplete. Before installing, consider whether the broad activation wording and the listed tool-install commands fit how you plan to use the skill.

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
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (6)

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

Autonomous Decision Making

Medium
Category
Excessive Agency
Confidence
75% confidence
Finding

Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.

Content

Scanner excerpt · SKILL.md (reported line 92)May include surrounding context.

md
When a tool is missing on the person's computer, name it, give the install
command for this system and wait. Install only after the person agrees. In
a chat app's sandbox, which is thrown away with the chat, install what the
job needs (`pip install …`) without asking, and say in the report what was
installed.

- macOS (Homebrew): `brew install ffmpeg imagemagick resvg qpdf poppler ghostscript img2pdf ocrmypdf tesseract-lang djvulibre pandoc sevenzip uchardet yq exiftool` and `brew install --cask libreoffice calibre`

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · SKILL.md (reported line 96)May include surrounding context.

md
installed.

- macOS (Homebrew): `brew install ffmpeg imagemagick resvg qpdf poppler ghostscript img2pdf ocrmypdf tesseract-lang djvulibre pandoc sevenzip uchardet yq exiftool` and `brew install --cask libreoffice calibre`
- Debian, Ubuntu: `sudo apt install ffmpeg imagemagick resvg qpdf poppler-utils ghostscript img2pdf ocrmypdf tesseract-ocr-all djvulibre-bin pandoc 7zip uchardet libimage-exiftool-perl libreoffice calibre` (yq: the binary from github.com/mikefarah/yq)
- Windows: `winget install Gyan.FFmpeg ImageMagick.ImageMagick JohnMacFarlane.Pandoc 7zip.7zip TheDocumentFoundation.LibreOffice calibre.calibre MikeFarah.yq`

In a chat app with code execution, work on the uploaded file inside the

Sudo/Root Execution

Medium
Category
Privilege Escalation
Confidence
70% confidence
Finding

Commands invoke sudo or root privileges. Verify this elevated access is necessary and justified.

Content

Scanner excerpt · scripts/av.py (reported line 37)May include surrounding context.

python
def need(tool):
    if not shutil.which(tool):
        fail(tool + " not found. Install ffmpeg (it brings ffprobe): "
             "macOS: brew install ffmpeg | Debian/Ubuntu: sudo apt install ffmpeg | "
             "Windows: winget install Gyan.FFmpeg", 4)

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/av.py (reported line 42)May include surrounding context.

python
def run(cmd):
    p = subprocess.run(cmd, stdin=subprocess.DEVNULL, capture_output=True, text=True)
    if p.returncode != 0:
        print("Failed: " + " ".join(cmd), file=sys.stderr)
        print(p.stderr.strip()[-3000:], file=sys.stderr)

The catch-all trigger “whenever a file has to ... be fixed” extends beyond file conversion.

Low
Category
Trigger Abuse
Confidence
88% confidence
Finding

“Be fixed” can refer to debugging source code, editing document content, or repairing application configuration, not just converting formats or encodings. The activation description does not exclude these unrelated repairs, so it could invoke the converter for tasks outside its intended scope.

Content

No source excerpt is available for this finding.