The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes shell scripts (launch.sh and launch.ps1) to coordinate its internal harness and manage the execution of creative tasks without modifying global system paths.
[EXTERNAL_DOWNLOADS]: Downloads official machine learning model weights for features like Face Tracking and Roto Brush from vendor-managed sources using system tools like curl.
[DYNAMIC_EXECUTION]: Includes a run_script functionality that enables the execution of JavaScript code within a specialized scripting model for After Effects-style automation, which is standard for high-end creative workflows.
[INDIRECT_PROMPT_INJECTION]: The skill has an attack surface for indirect injection as it processes complex external data formats such as PSD, SVG, and video footage.
Ingestion points: Untrusted data enters the context via mediaBrowser.import and file.import tools as documented in references/scenario.md and native-command-snapshot.json.
Boundary markers: The managed-execution.md documentation provides explicit warnings against passing raw natural language as native parameters.
Capability inventory: The skill possesses capabilities for native command execution and JavaScript scripting within the engine.
Sanitization: Creative tasks are structured as JSON plans and are subjected to strict schema validation (craft-command-plan/v1) and integrity checks before execution.

effectcraft-cli-footage