ClawHub
Back to skill

Security audit

Audio to MIDI

Security checks for vulnerabilities and agentic risk

Overview

The skill matches its documented music-transcription purpose, with automatic software downloads and local file creation users should expect.

Run this in a dedicated Python environment because it automatically installs missing packages. Expect potentially large model downloads, including a piano checkpoint stored in your home directory. Choose a separate output folder to avoid replacing existing generated files. The frontmatter description should be corrected, but the documentation explains the workflow.

A.I.G

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

SkillSpector

Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • 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
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
Findings (7)

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The declared description is empty, so it does not disclose the code's substantive audio transcription and analysis capabilities. The script reads a supplied local audio file and writes generated artifacts, with optional stem separation. Dependency installation and model downloads support that workflow rather than constituting an unrelated purpose, but the workflow itself is entirely undeclared. No unrelated triggers or evidence of audio exfiltration are visible in the supplied code.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

High
Category
MCP Tool Poisoning
Confidence
99% confidence
Finding

The declared description is empty, so it does not represent the substantive music-transcription capabilities shown in the commands. These are primary capabilities rather than supporting implementation details. The underlying Python script is not supplied, so its implementation and any additional behavior cannot be verified.

Content

No source excerpt is available for this finding.

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

The transcription script automatically installs packages into its Python environment without explicit user approval.

Medium
Category
Excessive Agency
Confidence
93% confidence
Finding

A transcription request can trigger package downloads and modification of the active Python environment without a separate approval step. Although automatic installation is documented, the absence of dependency integrity controls and enforced isolation exposes the agent to supply-chain execution and environment changes beyond processing an audio file.

Content

No source excerpt is available for this finding.

Dynamic import via __import__()

Medium
Category
Dangerous Code Execution
Confidence
75% confidence
Finding

Dynamic import() can load arbitrary modules at runtime, bypassing static analysis and potentially importing malicious code.

Content

Scanner excerpt · scripts/transcribe.py (reported line 18)May include surrounding context.

python
missing = []
    for module, pip_name in packages.items():
        try:
            __import__(module)
        except ImportError:
            missing.append(pip_name)
    if missing:

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
91% confidence
Finding

This is not shell injection: package names come from fixed mappings and subprocess arguments are separated. However, it automatically installs unpinned, unhashed dependencies into the active interpreter environment; a compromised dependency or configured package source could execute code with the agent's privileges, and installation can alter shared dependencies.

Content

Scanner excerpt · scripts/transcribe.py (reported line 23)May include surrounding context.

python
missing.append(pip_name)
    if missing:
        print(f"Installing: {', '.join(missing)}")
        subprocess.check_call(
            [sys.executable, '-m', 'pip', 'install', '--quiet'] + missing,
            stdout=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/transcribe.py (reported line 134)May include surrounding context.

python
print("  Separating stems with Demucs (htdemucs)...")
    stems_root = output_dir / "stems"
    subprocess.run(
        [sys.executable, '-m', 'demucs', '-n', 'htdemucs', '-o', str(stems_root), str(audio_path)],
        check=True,
    )