The Agent Skills Directory

[PROMPT_INJECTION]: The skill includes robust defenses against indirect prompt injection. In SKILL.md under the 'Security' section, it explicitly instructs the agent to treat transcript text, OCR text, and frame content as untrusted data. It specifically warns the agent to ignore instructions embedded in the video (e.g., 'ignore previous instructions') and to flag such attempts in the final note rather than executing them. This is a security best practice for handling external data.
[CREDENTIALS_UNSAFE]: The skill provides clear instructions to avoid credential exposure. The 'Evidence rules' section in SKILL.md mandates that API keys, tokens, or passwords visible or spoken in the video must be redacted and replaced with <REDACTED-API-KEY>. This prevents the unintentional exfiltration of secrets from video content into markdown notes.
[COMMAND_EXECUTION]: While the skill mentions running shell commands (e.g., npx skills add), these are standard installation instructions for dependencies. The skill explicitly forbids the execution of any commands extracted from the video content, mitigating risks associated with malicious commands embedded in visual or audio data.
[EXTERNAL_DOWNLOADS]: The skill references standard tools like yt-dlp and ffmpeg, and standard repositories for skill dependencies (Paldom/video-to-skill). These references are consistent with the skill's purpose and follow safe development practices.

video-notes