The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill references official Python and Node.js SDKs and documentation hosted on vendor-controlled platforms.
Evidence: Links to github.com/magichourhq/magic-hour-python, github.com/magichourhq/magic-hour-node, and docs.magichour.ai in references/setup.md and SKILL.md.
[DATA_EXFILTRATION]: The skill facilitates the transfer of user media files to the vendor's hosted environment for processing.
Evidence: Instructions for using get_upload_urls followed by a PUT request to upload media bytes to the Magic Hour API as described in SKILL.md.
Note: This is standard functionality for the service and includes explicit warnings against hardcoding credentials.
[COMMAND_EXECUTION]: Setup instructions include shell commands for configuring the Magic Hour MCP server within supported CLI environments.
Evidence: A codex mcp add command is provided in references/setup.md to help users configure their environment variables and server endpoint.
[INDIRECT_PROMPT_INJECTION]: The skill ingests user-provided media (images, audio, video) which serves as the primary input for its generation and editing capabilities.
Ingestion points: File paths for user assets such as assets.image_file_path and assets.audio_file_path in SKILL.md.
Boundary markers: Tool parameters are structured to separate creative prompts from asset references.
Capability inventory: Media generation tools (e.g., generate_image, create_video_from_image), file upload (via PUT to signed URLs), and result retrieval.
Sanitization: No explicit content sanitization for media file metadata is described.

magic-hour-media