The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute shell commands on the host machine to validate file accessibility and perform media analysis.
Evidence: Commands include head -c 1 "<path>" (Unix) and Get-Content -LiteralPath "<path>" -AsByteStream -TotalCount 1 (PowerShell) for file validation.
Evidence: Use of ffprobe and ffmpeg for analyzing video duration, aspect ratio, and scene transitions.
Evidence: Use of yt-dlp for downloading external content from user-provided URLs.
[EXTERNAL_DOWNLOADS]: The skill utilizes the yt-dlp utility to download video content from external social media URLs for use as pacing references.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from local files and external URLs, which could potentially contain malicious instructions intended to influence the agent's behavior during brand profile generation or post drafting.
Ingestion points: Local brand materials (bios, photos, documents) and external video URLs.
Boundary markers: The agent is instructed to present captured facts and a "brand brief" to the user for explicit approval before proceeding.
Capability inventory: Tools for brand creation, video generation, and post drafting; local command execution for media processing.
Sanitization: Instructions include constraints against generating person likenesses and guidelines to avoid inventing credentials or claims not explicitly provided by the user.
[DATA_EXFILTRATION]: The skill reads local files provided by the user and uploads their contents to the platform's infrastructure via HTTP PUT requests to presigned URLs. This behavior is the primary mechanism for populating the brand's knowledge base.

first-post