Back to skill

Security audit

短剧剪辑

Security checks for vulnerabilities and agentic risk

Overview

ClawScan rates this skill suspicious because it found substantial resource and file-write risks, while part of the artifact could not be fully inspected.

Consider avoiding installation until the media-processing limits, output-write boundaries, and tool scope have been reviewed. The scan is partial: it reports 18 of 27 components fully scanned, with additional files or checks only partially inspected, so it cannot establish that the skill is safe.

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
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
  • Prompt InjectionInstruction Override, Hidden Instructions, Exfiltration Commands
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Privilege EscalationExcessive Permissions, Sudo/Root Execution, Credential Access
Findings (17)

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 21)May include surrounding context.

md
剧集/<EP>/制作成果/` 中已有可读的视频文件(或关键帧图片与配音文件),且本机 `PATH` 中装有 `ffmpeg` 与 `ffprobe`(缺失时 `edit_tool.py` 直接报错退出)。同宽高比但分辨率或帧率不同的视频素材,`check` 列入提醒并由 `render` 自动缩放及转换帧率至交付规格;

Incomplete referenced artifact analysis

High
Category
analysis-evasion
Confidence
100% confidence
Finding

Referenced artifact was not completely inspected

Content

Scanner excerpt · SKILL.md (reported line 62)May include surrounding context.

md
命令行工具由 [edit_tool.py](scripts/edit_tool.py) 提供:

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

MCP server referenced without pinned version: 'npx remotion'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

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/edit_tool.py (reported line 1179)May include surrounding context.

python
def _has_ffmpeg_filter(ffmpeg: str, name: str) -> bool:
    result = subprocess.run(
        [ffmpeg, "-hide_banner", "-filters"],
        capture_output=True,
        text=True,

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/edit_tool.py (reported line 1204)May include surrounding context.

python
def probe_duration(media: Path) -> float:
    probe = _require("ffprobe")
    result = subprocess.run(
        [probe, "-v", "error", "-show_entries", "format=duration",
         "-of", "default=nw=1:nk=1", str(media)],
        capture_output=True, text=True, check=False,

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/edit_tool.py (reported line 1223)May include surrounding context.

python
def probe_stream(media: Path) -> dict[str, Any]:
    probe = _require("ffprobe")
    result = subprocess.run(
        [probe, "-v", "error", "-select_streams", "v:0", "-show_entries",
         "stream=width,height,r_frame_rate", "-show_entries", "format=duration",
         "-of", "json", str(media)],

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/edit_tool.py (reported line 1334)May include surrounding context.

python
"""(sample rate, channel layout) of the first audio stream; None when it has none."""

    probe = _require("ffprobe")
    result = subprocess.run(
        [probe, "-v", "error", "-select_streams", "a:0", "-show_entries",
         "stream=sample_rate,channels,channel_layout", "-of", "json", str(media)],
        capture_output=True, text=True, check=False,

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
97% confidence
Finding

The issue is resource exhaustion rather than shell injection: this decodes an entire audio stream and buffers all PCM output in memory before constructing another sample buffer. An untrusted long or highly compressed audio file can exhaust memory or monopolize execution, and voice inspection can reach this operation during both checking and rendering.

Content

Scanner excerpt · scripts/edit_tool.py (reported line 1504)May include surrounding context.

python
def _pcm_samples(ffmpeg: str, media: Path, rate: int, channels: int) -> Sequence[int]:
    """The media's sound as interleaved signed 16-bit samples. Raises when it cannot be read."""

    result = subprocess.run(
        [ffmpeg, "-v", "error", "-i", str(media), "-ac", str(channels), "-ar", str(rate),
         "-f", "s16le", "-"],
        capture_output=True, check=False,

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
89% confidence
Finding

This buffers the complete decoded RGB output without a byte limit or timeout. Video sampling reduces ordinary output volume but does not impose a total bound, and aspect-ratio-preserving scaling can produce an unexpectedly large second dimension for adversarial media.

Content

Scanner excerpt · scripts/edit_tool.py (reported line 2897)May include surrounding context.

python
def _sampled_stats(command: list[str]) -> Optional[ChannelStats]:
    result = subprocess.run(
        command + ["-pix_fmt", "rgb24", "-f", "rawvideo", "-"], capture_output=True, check=False
    )
    if result.returncode != 0 or len(result.stdout) < 3:

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/edit_tool.py (reported line 2930)May include surrounding context.

python
"""

    common = f"I={target}:TP={true_peak:g}:LRA=11"
    result = subprocess.run(
        [ffmpeg, "-hide_banner", "-nostats", "-i", str(media),
         "-af", f"loudnorm={common}:print_format=json", "-f", "null", "-"],
        capture_output=True, text=True, check=False,

MCP server referenced without pinned version: 'npx is'.

Medium
Category
MCP Rug Pull
Confidence
70% confidence
Finding

npx commands without a version suffix (e.g. @1.0.0) create a rug-pull risk if the upstream server is compromised and publishes a malicious update.

Content

No source excerpt is available for this finding.

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/edit_tool.py (reported line 3143)May include surrounding context.

python
encoding="utf-8",
    )
    overlay = output_root / "叠层.webm"
    result = subprocess.run(
        [npx, "remotion", "render", REMOTION_COMPOSITION, str(overlay),
         f"--props={props}", "--log=error",
         f"--concurrency={max(1, concurrency)}"],

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
89% confidence
Finding

The verification helper buffers all sampled RGB frames before computing aggregate colors, with no output-size or execution limit. A long segment or adversarial aspect ratio can cause excessive memory consumption despite the reduced sampling rate and nominal width.

Content

Scanner excerpt · scripts/edit_tool.py (reported line 3653)May include surrounding context.

python
def _segment_colour(ffmpeg: str, path: Path) -> Optional[tuple[float, float]]:
    """Mean luma and blue-red difference of one segment, on a 0-255 scale."""

    result = subprocess.run(
        [ffmpeg, "-v", "error", "-i", str(path), "-vf", "fps=2,scale=96:-1",
         "-pix_fmt", "rgb24", "-f", "rawvideo", "-"],
        capture_output=True,

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/edit_tool.py (reported line 3711)May include surrounding context.

python
else "未测(剪辑单没有声明目标时长)"
        ),
    }
    result = subprocess.run(
        [ffmpeg, "-hide_banner", "-nostats", "-i", str(final),
         "-af", "loudnorm=print_format=json", "-f", "null", "-"],
        capture_output=True, text=True, check=False,

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
98% confidence
Finding

This decodes every frame into captured stdout and then copies the output into a list of individual frame buffers. Although frame dimensions are fixed, duration and frame count are not bounded, so an oversized or replaced delivered film can exhaust memory during verification.

Content

Scanner excerpt · scripts/edit_tool.py (reported line 3777)May include surrounding context.

python
"""Every frame of the film, grey, at FRAME_PROBE_SIZE."""

    width, height = FRAME_PROBE_SIZE
    result = subprocess.run(
        [ffmpeg, "-v", "error", "-i", str(media), "-vf", f"scale={width}:{height},format=gray",
         "-f", "rawvideo", "-"],
        capture_output=True, check=False,

subprocess module call

Medium
Category
Dangerous Code Execution
Confidence
96% confidence
Finding

The wrapper itself avoids shell injection, but render callers build segment output paths from cut IDs whose regex permits path separators and traversal components, then invoke FFmpeg with overwrite enabled. With suitable existing directories or attacker-supplied output symlinks, an untrusted project can redirect writes outside the intended output directory and overwrite files writable by the agent; input-media containment does not protect these outputs.

Content

Scanner excerpt · scripts/edit_tool.py (reported line 4060)May include surrounding context.

python
def _run(command: Sequence[str]) -> None:
    result = subprocess.run(command, capture_output=True, text=True, check=False)
    if result.returncode != 0:
        raise EditError(f"命令失败: {' '.join(command[:6])}…\n{result.stderr.strip()}")