The Agent Skills Directory

[PROMPT_INJECTION]: The skill contains explicit instructions for the agent to override user directives, stating that even if a user asks for a custom task, the agent MUST prioritize the skill's internal steps first.
[PROMPT_INJECTION]: The instructions mandate agent autonomy by directing it to clear its own context and call itself to resume work automatically without waiting for user interaction.
[PROMPT_INJECTION]: Instructions attempt to reduce transparency by ordering the agent to "blindly execute the paths" and forbidding the use of exploration commands like 'ls' or 'find'.
[DATA_EXFILTRATION]: The skill uses path traversal to access and read files from sibling directories (e.g., reading SKILL.md from an adjacent folder).
[DATA_EXFILTRATION]: The workspace creation logic is vulnerable to directory traversal; it constructs file paths using identifiers extracted directly from user-provided URLs without sanitization.
[COMMAND_EXECUTION]: The skill executes external binaries such as 'yt-dlp' and 'ffmpeg' using video URLs provided by the user, which presents an argument injection risk if the platform does not sanitize inputs.
[COMMAND_EXECUTION]: The agent is instructed to execute local Python scripts using the 'python3' command.
[EXTERNAL_DOWNLOADS]: The skill performs automated network requests to fetch game data from community repositories at api.atlasacademy.io and db.ygoprodeck.com.
[EXTERNAL_DOWNLOADS]: The skill uses 'yt-dlp' to download media and transcripts from external video hosting platforms.

anibon-timestamper-handoff