The Agent Skills Directory

[COMMAND_EXECUTION]: The skill orchestrates media tasks by executing several internal Python scripts (such as workflow.py, bootstrap.py, and domain_commands.py) as subprocesses. These scripts interact with native CLI tools to perform specialized editing operations.
[EXTERNAL_DOWNLOADS]: For speech-to-text capabilities, the skill downloads Whisper AI models from Hugging Face. These downloads are constrained to specific content revisions and are verified against known hashes to prevent tampering.
[INDIRECT_PROMPT_INJECTION]: The skill processes complex JSON-based workflow plans and brand briefs which dictate the sequence of operations for the agent.
Ingestion points: Processes user-supplied JSON files (e.g., brief.json, brand-campaign.json) and external assets like images and audio.
Boundary markers: Implements strict versioning for schemas (e.g., craft-skill-workflow/v1) and uses "frozen" execution plans to prevent modification during runtime.
Capability inventory: Possesses capabilities to perform file system operations, execute subprocesses, and initiate network requests via its orchestration scripts.
Sanitization: Incorporates extensive validation mechanisms, including SHA-256 checksum verification for all inputs/outputs, path sanitization to prevent directory traversal, and strict distribution locks for all dependencies.

artcraft-cli-execute