Back to skill

Security audit

Vlog Auto Edit

Security checks for vulnerabilities and agentic risk

Overview

This is a coherent AI video-editing skill, but it needs review because it uploads personal media frames to vision APIs and includes unsafe command and dashboard-generation patterns.

Use this only in a dedicated project folder with media you are comfortable processing. Require the agent to show commands before running them, use a virtual environment with pinned dependencies, avoid the shell=True highlight snippet, and confirm which frames or transcripts will be sent to any vision API. Treat generated dashboard HTML as private project data, especially if filenames, speech, or notes may contain sensitive information.

Vulnerability Patterns
  • Insecure DependenciesIntroduces malicious components through unsafe dependency sources
  • 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
Findings (3)

T09 · Insecure Skill Coding Practices

Error
Location
scripts/gen_dashboard.py:289
Finding

Stored HTML and JavaScript Injection in Generated Dashboard

Content
View full analysis
0) {{ usageHtml = c.usage.map(u => '
📎 ' + u.section + ' [' + u.start + 's→' + u.end + 's] ' + u.note + '
' ).join(''); }} else {{ usageHtml = '未在成品中使用'; }} let speechHtml = ''; if (c.speech.length > 0) {{ speechHtml = c.speech.map(s => '
💬 ' + s + '
').join(''); }} detail.innerHTML = '

📹 ' + c.file + '

' + '
' + '
' + thumbs + '
' + '
' + '
时长:' + c.dur.toFixed(1) + '秒
' + '
画面:' + c.visual + '
' + (speechHtml ? '
语音内容:' + speechHtml + '
' : '') + '
使用情况:
' + usageHtml + '
' + '
' + '
'; ``` ### Technical Analysis The dashboard embeds data derived from analysis and edit-plan JSON files directly into an executable `
Remediation
View remediation
... ``` Before embedding it, escape at least `<`, `>`, `&`, U+2028, and U+2029. Parse it through the element's `textContent`. 2. Replace dynamic `innerHTML` construction with safe DOM APIs: ```javascript const name = document.createElement('h3'); name.textContent = c.file; detail.replaceChildren(name); ``` 3. Use `textContent` for filenames, transcripts, notes, descriptions, and section names. Only use `innerHTML` for fixed, developer-controlled markup. 4. If HTML templates remain necessary, apply a well-reviewed context-aware sanitizer and keep untrusted values separate from markup. 5. Validate filenames and generated relative paths. Reject control characters, traversal sequences, and characters that are unnecessary for supported media filenames. 6. Add regression tests containing payloads such as closing script sequences, image error handlers, quotes, angle brackets, and multiline transcript content. Verify that they are displayed as text and never executed. ]]>

T09 · Insecure Skill Coding Practices

Error
Location
SKILL.md:530
Finding

Shell Command Injection in the Documented Highlight Rendering Workflow

Content
View full analysis
Remediation
View remediation

T08 · Insecure Dependencies

Warning
Location
SKILL.md:47
Finding

Unpinned Runtime Installation of Third-Party Dependencies

Content
View full analysis
/dev/null && echo "scenedetect is installed" || pip install scenedetect ``` ### Technical Analysis The Skill instructs the Agent to install third-party packages at runtime without: - Exact version constraints - Cryptographic hashes - A reviewed lock file - An explicitly trusted package index - Verification of package provenance - User confirmation before optional installations Package installation can execute package build and installation logic. Because unconstrained names resolve to the latest package versions available through the active pip configuration, the effective dependency code can change after the Skill has been audited. The package names shown are consistent with the declared video-editing functionality, and no typosquatted package name was identified. The issue is the unsafe installation process rather than proof that the named packages are malicious. The recommendation to install directly into the user environment also broadens the effect of dependency conflicts and compromises beyond the current project. ### Attack Path 1. The Agent checks for a dependency and determines that it is missing. 2. It runs an unpinned `pip install` command as instructed. 3. Pip resolves packages and transitive dependencies using the currently configured index or mirror. 4. A compromised release, malicious mirror response, dependency-confusion condition, or later unsafe package version is selected. 5. Package installation or subsequent import executes the compromised code with the Agent's user-level privil ...[truncated 738 chars]
Remediation
View remediation
Vulnerability Patterns
  • Data ExfiltrationExternal Transmission, Env Variable Harvesting, File System Enumeration
  • Tool MisuseTool Parameter Abuse, Chaining Abuse, Unsafe Defaults
  • Trigger AbuseOverly Broad Trigger, Shadow Command Trigger, Keyword Baiting Trigger
  • Behavioral ASTexec() Call, eval() Call, Dynamic Import
  • MCP Least PrivilegeUnderdeclared Capability, Wildcard Permission, Missing Permission Declaration
Findings (21)

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
99% confidence
Finding

This code builds a shell command with interpolated file paths and timing values, then executes it with shell=True. If any part of the input such as a filename is attacker-controlled, shell metacharacters can break out of the intended ffmpeg command and execute arbitrary OS commands.

Content

Scanner excerpt · SKILL.md (reported line 545)May include surrounding context.

md
f'pad=1920:1080:(ow-iw)/2:(oh-ih)/2" ' \
          f'-c:v libx264 -preset ultrafast -crf 23 -pix_fmt yuv420p ' \
          f'-an highlights/hl_{i}.mp4'
    subprocess.run(cmd, shell=True)

# concat
with open('highlights/concat.txt', 'w') as f:

Tool Parameter Abuse

High
Category
Tool Misuse
Confidence
87% confidence
Finding

Although this invocation uses a mostly static shell string, it still runs via shell=True and relies on a concat file generated earlier. In this workflow, that is less severe than the previous finding, but it preserves unnecessary shell interpretation and compounds risk if surrounding file paths or working directory state are influenced by untrusted input.

Content

Scanner excerpt · SKILL.md (reported line 552)May include surrounding context.

for i in range(1, len(highlights)+1): f.write(f"file 'hl_{i}.mp4'\n")

subprocess.run('ffmpeg -y -f concat -safe 0 -i highlights/concat.txt ' '-c copy highlights/montage.mp4', shell=True)

text

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The README explicitly instructs use of a third-party visual understanding API to analyze user video content, but does not clearly warn that frames or clips may be transmitted off-device to external providers. For a vlog-editing skill, the content can contain faces, voices, locations, license plates, and other sensitive personal data, so the omission creates a real privacy and data-handling risk rather than a purely informational issue.

Content

No source excerpt is available for this finding.

Undeclared Tool Scope

Medium
Category
MCP Least Privilege
Confidence
95% confidence
Finding

The skill clearly instructs the agent to write files and execute shell commands, but it does not declare any explicit tool scope such as permissions or allowed-tools. That creates an authorization ambiguity where a host agent may permit powerful operations implicitly, increasing the chance of unexpected file modification or command execution beyond what the user intended.

Content

No source excerpt is available for this finding.

Vague Triggers

Medium
Category
Not specified by scanner
Confidence
93% confidence
Finding

The trigger phrases include very broad requests like '帮我剪个视频' and '自动剪辑', which can match common user requests outside a clearly bounded workflow. Overbroad activation can cause the skill to engage in contexts where users did not intend a shell-heavy, file-writing, API-uploading automation pipeline to run.

Content

No source excerpt is available for this finding.

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
97% confidence
Finding

The skill directs the agent to extract local video frames and send them to an external vision API, but it does not require explicit user notice or consent for this data transfer. Travel and personal vlog footage can contain faces, locations, documents, or other sensitive content, so silent upload creates a real privacy and data-sharing risk.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
89% confidence
Finding

The entire markdown file presents the skill documentation only in Chinese, with no indication that users can choose another language or that the skill is intentionally limited to a Chinese-speaking audience. Under the stated policy, forcing a specific language without user opt-in or clear justification is a natural-language policy violation.

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/gen_dashboard.py (reported line 48)May include surrounding context.

python
"""执行 ffmpeg 命令"""
    cmd = ["ffmpeg", "-y"] + args
    kw = {"capture_output": True} if quiet else {}
    return subprocess.run(cmd, **kw)


# ── 数据加载与交叉匹配 ──────────────────────────────────

Missing User Warnings

Medium
Category
Not specified by scanner
Confidence
91% confidence
Finding

The script collects speech transcripts and visual descriptions from source media and embeds them directly into a browsable HTML dashboard without any privacy notice, minimization, or access-control consideration. In a vlog workflow, raw footage often contains sensitive incidental speech, names, locations, or bystanders, so generating a shareable artifact can unintentionally expose personal data.

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/gen_dashboard.py (reported line 224)May include surrounding context.

python
return []

    # 获取视频时长
    probe = subprocess.run(
        ["ffprobe", "-v", "quiet", "-print_format", "json", "-show_format", video_path],
        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/gen_dashboard.py (reported line 646)May include surrounding context.

python
if not video_path or not os.path.exists(video_path):
        return None

    probe = subprocess.run(
        ["ffprobe", "-v", "quiet", "-print_format", "json",
         "-show_format", "-show_streams", video_path],
        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/gen_storyboard.py (reported line 58)May include surrounding context.

python
"-vf", "scale=480:-1",
                frame_path,
            ]
            subprocess.run(cmd, capture_output=True)
            status = "✓" if os.path.exists(frame_path) else "✗"
            print(f"  {status} frame_{clip_idx:03d}: {fname} @{mid:.1f}s")

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/gen_dashboard.py (reported line 767)May include surrounding context.

python
# macOS 自动打开
    if sys.platform == "darwin":
        subprocess.run(["open", html_path])


if __name__ == "__main__":

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/gen_storyboard.py (reported line 300)May include surrounding context.

python
# macOS 自动打开
    if sys.platform == "darwin":
        subprocess.run(["open", html_path])


if __name__ == "__main__":

Natural-Language Policy Violations

Medium
Category
Not specified by scanner
Confidence
96% confidence
Finding

The template's instructions are entirely in Chinese and prescribe Chinese field descriptions and response guidance, effectively constraining the interaction language. The file does not offer a language choice or explain that this is a region-specific or opt-in locale requirement, which matches the language/locale policy violation criteria.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
84% confidence
Finding

The quick-start text says the agent will automatically complete installation and configuration steps, which can lead users to approve package installation and environment changes without understanding that their system may be modified. In this skill context the actions are ordinary setup steps, but the lack of an explicit warning reduces informed consent and could normalize unsafe autonomous changes by an agent.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
93% confidence
Finding

This JSON file contains user-facing natural-language descriptions, transcripts, and notes exclusively in Chinese, including fields that appear intended for downstream consumption. Because the file does not provide any language choice, opt-in, or justification for a Chinese-only locale, it may violate the language/locale policy requirement.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
90% confidence
Finding

This JSON file contains all user-facing titles, section names, descriptions, notes, subtitles, and editing notes in Chinese, indicating a fixed language output. The file does not offer any language or locale opt-in, nor does it document that the skill is intentionally region-specific.

Content

No source excerpt is available for this finding.

Missing User Warnings

Low
Category
Not specified by scanner
Confidence
83% confidence
Finding

The script creates output directories and writes thumbnails, QC frames, and an HTML dashboard to disk, including overwrite behavior in ffmpeg via the '-y' flag. Although these writes are part of the tool's purpose, there is no explicit warning that existing derived files may be created or replaced during execution.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
96% confidence
Finding

The HTML template hard-codes lang="zh-CN", which enforces a specific language/locale setting regardless of user preference. The file does not indicate that this is an optional or region-specific behavior, so it violates the language-choice policy for natural-language behavior.

Content

No source excerpt is available for this finding.

Natural-Language Policy Violations

Low
Category
Not specified by scanner
Confidence
95% confidence
Finding

This code emits an HTML document with lang="zh-CN", and the surrounding generated labels and headings are Chinese-only throughout the file. Under the policy rule, forcing a specific language/locale without opt-in is a natural-language policy violation unless it is explicitly justified or configurable.

Content

No source excerpt is available for this finding.