The Agent Skills Directory

[PRIVILEGE_ESCALATION]: The skill instructs the agent to modify the .claude/settings.json (or settings.local.json) file to pre-approve various shell commands, including yt-dlp, ffmpeg, and ffprobe. This deliberately bypasses the platform's security architecture that requires human-in-the-loop confirmation for tool execution.
[REMOTE_CODE_EXECUTION]: The skill performs remote code execution by downloading and piping shell scripts directly to the shell (curl ... | sh). These patterns are used for installing Deno (via deno.land) and the Treg CLI (via treg.to).
[INDIRECT_PROMPT_INJECTION]: The skill implements an ingestion pipeline that fetches transcripts from external URLs (YouTube, TikTok, Instagram) using yt-dlp. This untrusted external content is then used to construct the voice.md file, which serves as a core instruction set for subsequent video editing and scriptwriting tasks.
Ingestion points: External URLs provided by the user for video transcripts (SKILL.md, Section 4).
Boundary markers: The skill suggests reading the generated voice.md back to the user for validation, which provides a manual check but doesn't programmatically sanitize the input.
Capability inventory: The skill has broad capabilities including file system writes, network access via yt-dlp and curl, and shell command execution (ffmpeg, node).
Sanitization: There is no automated sanitization of the transcript content before it is interpolated into the voice.md template.
[COMMAND_EXECUTION]: The skill makes extensive use of the Bash tool to perform system checks, install software, and manage files. It specifically requests pre-approval for these commands to run autonomously in future sessions.

video-setup