The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The scripts seedance.py and send_video_feishu.py contain hardcoded Feishu API credentials, including FEISHU_APP_ID ('cli_a9005be686f89bd3') and FEISHU_APP_SECRET ('UvUMxhBGbejzqbeuR5BQqeZnjCrnsCiC'). Hardcoding secrets in skill files exposes them to anyone with access to the skill source code.
[COMMAND_EXECUTION]: The script seedance.py uses os.system(f'open "{filepath}"') to launch video files on macOS. The filepath variable includes the task_id, which is a string returned by the Volcengine Ark API. If the remote API returns a malicious payload (e.g., containing shell metacharacters like backticks or subshell expansions), it could result in arbitrary command execution on the user's system.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection vulnerabilities due to the following evidence chain:
Ingestion points: The skill ingests untrusted data from user prompts (--prompt argument) and from remote API responses (task_id, video_url).
Boundary markers: There are no boundary markers or sanitization logic to delimit user-provided text or API-returned strings before they are used in shell commands or interpolated into API requests.
Capability inventory: The skill has the capability to write to the file system, make network requests to external APIs, and execute shell commands via os.system.
Sanitization: No escaping or validation is performed on the task_id before it is passed to the shell environment.

seedance-video