Back to skill

Security audit

video-content-pipeline

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly a coherent video-generation helper, but it includes an optional paid API-key scene generator that is not declared in the manifest permissions.

Review before installing. The free pipeline sends prompts to Pollinations and narration to Microsoft Edge TTS as disclosed. Do not use the paid scene_gen.py feature unless you intentionally set X402_API_KEY, understand the endpoint configured by X402_BASE, and accept possible charges and topic/API-key sharing. On shared machines, avoid running the compositor in a sensitive environment until its /tmp file handling is hardened.

Vulnerability Patterns
  • Excessive AgencyUnrestricted Tool Access, Autonomous Decision Making, Scope Creep
  • Output HandlingUnvalidated Output Injection, Cross-Context Output, Unbounded Output
  • MCP Tool PoisoningHidden Instructions, Unicode Deception, Parameter Description Injection
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
Findings (8)

Tp4

High
Category
MCP Tool Poisoning
Confidence
98% confidence
Finding

The top-level description discloses only Pollinations and Microsoft Edge TTS, but the documentation later introduces an additional paid scene-generation API that sends user topic data externally and may charge the user's key. This mismatch can cause users or orchestration systems to trust the skill under incomplete assumptions about data egress, billing exposure, and required permissions.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The documented scope claims a full pipeline, but also includes an extra paid external capability outside the declared manifest scope. This undermines informed consent and can lead to accidental external data sharing or unexpected behavior by agents that rely on the manifest as the source of truth.

Content

No source excerpt is available for this finding.

Scope Creep

High
Category
Not specified by scanner
Confidence
97% confidence
Finding

The paid scene-generation workflow implies outbound network access to an additional service that is not listed in the manifest permissions. Undeclared network destinations are dangerous because they bypass policy review, can exfiltrate user prompts or secrets, and may trigger unapproved financial transactions.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

This file introduces a paid external x402 scene-generation API that is not described in the stated skill purpose, which only mentions Pollinations image generation, Edge TTS, and composition. That mismatch is dangerous because users may unknowingly trigger billable external calls and send content to a third party outside the declared pipeline.

Content

No source excerpt is available for this finding.

Context-Inappropriate Capability

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The code handles a spending-capable API key and performs paid transactions, yet this financial capability is not justified by the skill’s declared purpose. In an agent-skill context, unexpected monetized actions materially increase risk because a user or orchestrator may execute the script without understanding it can incur charges or expose a billable credential to an external service.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
82% confidence
Finding

The code writes to a fixed path (/tmp/concat.txt) without creating a unique temporary file securely. On multi-user systems, an attacker could pre-create this path as a symlink or race its creation, potentially causing overwrites of unintended files or manipulating the concat input used by ffmpeg.

Content

No source excerpt is available for this finding.

Unvalidated Output Injection

High
Category
Output Handling
Confidence
90% confidence
Finding

The script writes intermediate output to predictable filenames in /tmp based on scene IDs, e.g. /tmp/scene_XX.mp4, without securely creating them. An attacker on the same host could exploit symlink or race conditions to overwrite arbitrary files accessible to the process, or poison the intermediate artifacts that are later concatenated.

Content

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

python
if not (os.path.exists(img) and os.path.exists(aud)):
            print(f"⚠️ missing scene {sc['id']}"); continue
        part = f"/tmp/scene_{sc['id']:02d}.mp4"
        subprocess.run([
            "ffmpeg", "-y", "-loop", "1", "-i", img, "-i", aud,
            "-c:v", "libx264", "-tune", "stillimage", "-c:a", "aac",
            "-shortest", "-pix_fmt", "yuv420p", part

Unvalidated Output Injection

High
Category
Output Handling
Confidence
91% confidence
Finding

The final ffmpeg invocation accepts an unvalidated output path (out) directly from argv and also processes a concat file with -safe 0, reducing ffmpeg’s path restrictions. While this is not shell injection, it can enable arbitrary file overwrite in locations writable by the process and makes concat handling more dangerous if temporary-file manipulation occurs.

Content

Scanner excerpt · scripts/compositor.py (reported line 30)May include surrounding context.

python
# Escape enkelt-anførselstegn i filstier (ffmpeg concat-injektion fix 20/8)
    lines = ["file '" + p.replace("'", "'\\''") + "'\n" for p in parts]
    open(concat_file, "w").write("".join(lines))
    subprocess.run(["ffmpeg", "-y", "-f", "concat", "-safe", "0", "-i", concat_file,
                    "-c", "copy", out], capture_output=True)
    print(f"✅ Video: {out}")