Back to skill

Security audit

Video Rough Cut

Security checks for vulnerabilities and agentic risk

Overview

The skill is mostly coherent, but its helper script can send the selected video to any API URL, not just the local service the documentation describes.

Review this skill before installing if your videos may be private or business-sensitive. Use it only with a trusted local B-Roll Studio service, avoid passing a non-local --base-url, and prefer a version that enforces localhost-only uploads or requires explicit opt-in for remote processing.

Vulnerability Patterns
  • Insecure Skill Coding PracticesFinds exploitable flaws such as hardcoded secrets or command injection
  • 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
Findings (1)

T09 · Insecure Skill Coding Practices

Warning
Location
run_rough_cut.py:31
Finding

Unrestricted API destination can disclose selected local video files

Content
View full analysis
tuple[bytes, str]: boundary = "----CodexRoughCutBoundary" chunks: list[bytes] = [] for key, value in data.items(): chunks.extend([ f"--{boundary}\r\n".encode(), f'Content-Disposition: form-data; name="{key}"\r\n\r\n'.encode(), str(value).encode(), b"\r\n", ]) mime = mimetypes.guess_type(file_path.name)[0] or "application/octet-stream" file_bytes = file_path.read_bytes() chunks.extend([ f"--{boundary}\r\n".encode(), f'Content-Disposition: form-data; name="file"; filename="{file_path.name}"\r\n'.encode(), f"Content-Type: {mime}\r\n\r\n".encode(), file_bytes, b"\r\n", f"--{boundary}--\r\n".encode(), ]) return b"".join(chunks), boundary ``` ```python def submit_job(args: argparse.Namespace) -> dict: video = Path(args.video).expanduser().resolve() if not video.exists(): raise FileNotFoundError(f"Video not found: {video}") data = { "name": args.name or video.stem, "platform_preset": args.platform_preset, "remove_pauses": bool_string(args.remove_pauses), "remove_breaths": bool_string(args.remove_breaths), "adjust_brightness": bool_string(args.adjust_brightness), "apply_beauty": bool_string(args.beauty_mode != "off"), "trim_head_tail": bool_string(args.trim_head_tail), "auto_center": bool_string(args.auto_center), "stabilize": bool_string(args.stabilize), "denoise_audio": bool_string(args.denoise_audio), ...[truncated 2862 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • 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 (5)

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
91% confidence
Finding

The skill instructs the agent to read local files and send the user’s video to a local HTTP service, but it does not declare any explicit tool scope or allowed-tools boundary. That creates an authorization and transparency gap: the runtime may permit broader file or network access than users expect, and reviewers cannot easily verify the minimum required privileges.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The description says the skill 'uploads one raw video' and exports a cleaned draft, but it does not present this as an explicit user-facing warning about data movement to a local service and creation/download of modified output. For media workflows, the video may contain sensitive personal or business content, so failing to clearly disclose transfer and output behavior can lead to unintended exposure or processing without informed consent.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The file includes Chinese-only clutter examples and mixed-language guidance such as 321走 / 试音 / 收镜头动作, 正文, and 闷 without indicating that the skill is limited to Chinese-language workflows or offering an opt-in language choice. Under the policy, forcing or assuming a specific language/locale without user opt-in is a natural-language policy violation.

Content

No source excerpt is available for this finding.

Description-Behavior Mismatch

Medium
Category
Not specified by scanner
Confidence
95% confidence
Finding

The skill is described as using a local rough-cut pipeline, but the script accepts an arbitrary --base-url and then uploads the full input video and processing parameters to that endpoint. In an agent setting, this can silently exfiltrate sensitive local media to a remote server if a prompt, wrapper, or caller supplies a non-local URL, and there is no validation, allowlist, or user confirmation to preserve the 'local-only' trust boundary implied by the skill metadata.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
82% confidence
Finding

The trigger list mixes Chinese and English phrases, but the file does not say whether the skill is intended only for bilingual contexts or whether users can choose language behavior. This can be a natural-language policy issue when language expectations are imposed implicitly rather than offered as a choice or justified as region-specific.

Content

No source excerpt is available for this finding.