The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads the filmcraft-cli binary and a desktop application (FilmCraft.app) from GitHub repositories under the full-aigc-skills and storytold organizations. Integrity is maintained by validating these files against SHA256 checksums defined in runtime.lock.json and desktop.lock.json before any execution occurs.
[COMMAND_EXECUTION]: Python's subprocess module is used to execute the downloaded binaries and various system utilities (such as hdiutil, codesign, and lipo). These operations are protected by a kernel-level sandbox (sandbox-exec) and strict validation of command-line arguments to prevent unauthorized access or privilege escalation.
[DYNAMIC_EXECUTION]: The skill utilizes importlib to dynamically load internal scripts from its own scripts/ directory. This is used for modular functionality and is constrained to files within the skill's own directory structure.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data including JSON editing plans and media file metadata. This represents a potential vulnerability surface where malicious instructions could be embedded in data.
Ingestion points: workflow.py and commands.py read user-supplied JSON plans and media headers.
Boundary markers: The skill relies on structured JSON parsing and does not use explicit NL delimiters.
Capability inventory: File system writes, subprocess execution, and network access (restricted to maintenance tasks) are available in workflow.py and cli.py.
Sanitization: The skill implements custom logic to detect duplicate JSON keys and uses a comprehensive path-validation policy to ensure files are only accessed within authorized roots.

filmcraft-cli-multicam