The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructions in SKILL.md explicitly guide the agent to prompt the user for their ACCESS_KEY_ID and SECRET_ACCESS_KEY and then set them as session-scoped environment variables using export. This involves the direct handling of sensitive cloud credentials within the chat interface.
[EXTERNAL_DOWNLOADS]: The skill installs dependencies from scripts/requirements.txt, including pandas==2.3.3. This specific version is not currently available on the public PyPI registry, which will result in installation errors. Furthermore, the skill directs the agent to download video files from user-provided URLs using curl to the /tmp directory for processing.
[COMMAND_EXECUTION]: The skill's operation depends on executing several local Python scripts (plan.py, upload.py, subtitler.py, upgrade.py) and shell commands like mkdir, curl, and file to manage the video subtitle workflow and file validation.
[DATA_EXFILTRATION]: As part of its primary function, the skill reads local video assets and uploads them to ByteDance's infrastructure at icp.volcengineapi.com. While this is the intended purpose, it involves the transmission of potentially private media files to external servers.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external sources, such as video files from URLs and user-defined captions in JSON or SRT formats. Although it uses the file utility to perform basic media type checks, these inputs represent an attack surface where malformed files or embedded instructions could attempt to influence the agent's behavior.

byted-kickart-video-subtitler