pptx-creator

================================================================================

🔴 VERDICT: CRITICAL

This skill has a critical command execution vulnerability. It uses subprocess.run to execute another skill (nano-banana-pro) for image generation, passing user-controlled input directly as a command-line argument (--prompt). A malicious user could craft this prompt to inject arbitrary shell commands, leading to remote code execution on the system running the agent. Additionally, this constitutes a prompt injection vector for the nano-banana-pro skill itself. The skill also relies on an unverified external skill (nano-banana-pro) and contains placeholder code for CRM API access that could expose credentials if activated.

Total Findings: 4

🔴 CRITICAL Findings: • Command Execution via User-Controlled Subprocess Argument

• scripts/create_pptx.py
• Line 190: subprocess.run(["uv", "run", str(script_path), "--prompt", prompt, "--filename", output_path, "--resolution", "1K"], ...)

🔴 HIGH Findings: • Prompt Injection to External Skill

• scripts/create_pptx.py
• Line 185: prompt = src.split("generate:", 1)[1].strip() (User-controlled prompt passed to nano-banana-pro via subprocess)

🟡 MEDIUM Findings: • Unverifiable External Skill Dependency

• scripts/create_pptx.py
• Line 181: script_path = SKILL_DIR.parent / "nano-banana-pro" / "scripts" / "generate_image.py" (Relies on nano-banana-pro skill, which is not audited here)

🔵 LOW Findings: • Potential Credential Exposure (Placeholder)

• scripts/create_pptx.py
• Line 208: api_url = os.environ.get("TWENTY_API_URL", "") and Line 209: api_token = os.environ.get("TWENTY_API_TOKEN", "") (Reads sensitive environment variables for a currently inactive CRM API integration)

================================================================================