Security audit
Security checks for vulnerabilities and agentic risk
The skill is a disclosed short-video download and transcription workflow, but users should be aware it writes local media/text outputs and depends on other local scripts/skills.
Install only if you intend to process Douyin/Bilibili or local media into audio and text artifacts. Expect network downloads, local output folders, and Chinese-first transcription behavior; review the required sibling skills/scripts and choose a non-sensitive output directory before running.
Code accesses credential files (SSH keys, AWS credentials, etc.). This could indicate credential theft attempts.
venv/
ENV/
env/
.env
# Logs
logs/*.log
A description-behavior mismatch is security-relevant because users and orchestrators may authorize the skill for one workflow while the implementation actually invokes different components or fails to enforce the claimed end-to-end data flow. In this context, the skill handles external URLs, local filesystem paths, and multiple script handoffs, so inaccurate declarations can conceal unexpected processing, data placement, or execution behavior.
The README promotes a one-click workflow that downloads videos, extracts audio, and writes multiple derived artifacts such as MP3s, vocals, transcripts, corrected text, and final text to disk, but it does not clearly warn users up front about these persistent side effects. In a chat-triggered agent context, this can cause unexpected local storage use, privacy exposure, and inadvertent retention of copyrighted or sensitive content, especially when processing user-supplied URLs or local media folders.
The skill description states that transcription and correction are Chinese-first and presents Chinese text processing as the default behavior without any visible language selection or opt-in. This is risky because users may unknowingly process non-Chinese content with the wrong models, causing inaccurate transcripts, corruption of outputs, and possible mishandling of multilingual or sensitive material.
The skill advertises executable workflow steps that invoke Python CLI scripts and therefore has shell/code-execution capability, but it declares no explicit tool scope or allowed-tools boundary. In a composite skill that downloads external content and writes files, missing capability scoping increases the chance of unintended command execution paths or over-broad agent permissions.
Overly broad trigger phrases can cause the skill to activate on ordinary user language, leading to unintended downloading of external videos, processing of local media, and filesystem writes. Because this skill performs a multi-step automated pipeline with external inputs, accidental invocation materially increases operational and privacy risk.
The skill performs network retrieval of third-party video content and writes outputs into local directories, but the description does not clearly warn the user about these side effects. Without transparent notice and confirmation, users may unknowingly authorize external-content handling, storage of derived artifacts, or processing of sensitive local files.
The top-level documentation names a different skill entirely, while the rest of the file implements a six-step short-video content replication pipeline. This is an active documentation contradiction, not just missing detail, and can mislead reviewers or orchestrators about the skill's true intent.
subprocess module calls execute external commands. Without careful input validation, this enables command injection.
"""统一执行子脚本，带友好日志、错误捕获和输出显示"""
try:
console.print(f"\n[bold cyan]▶ Step {step_name}[/bold cyan]")
result = subprocess.run(
cmd,
check=True,
capture_output=True,
The workflow hard-codes the transcription language to "zh" when invoking the audio transcription step. This enforces a specific language/locale behavior without offering the user a choice or documenting a justified region-specific restriction, which matches the policy-violation criteria.
The activation description and trigger phrases are entirely specified in Chinese, and the workflow centers on Chinese-language phrasing without stating whether other user languages are supported. This can constitute a language policy issue when a skill implicitly requires a specific language without opt-in or an explicit locale justification.