The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill makes extensive use of the adobe_execute_script tool to run dynamically generated ExtendScript (ES3) directly within the Adobe After Effects environment. This is a core functionality that allows the agent to programmatically manipulate compositions, layers, and effects. It also utilizes $.evalFile to load and execute local script libraries from computed or absolute paths.
[EXTERNAL_DOWNLOADS]: Several tools, including adobe_place_element, adobe_install_fonts, and adobe_import_generation, are designed to download and install external assets such as overlays, templates, fonts, and AI-generated footage from the vendor's 'Mira Elements' library and related services.
[DATA_EXFILTRATION]: The adobe_render tool includes a feature to optionally upload rendered video files to the vendor's 'Mira library'. While this involves sending user data to an external service, it is a declared and integral feature of the Mira workflow.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection as it ingests and processes untrusted data from existing user projects and external video references.
Ingestion points: Project data is read via adobe_get_project (including layer names, source text, and markers) and external videos are analyzed via adobe_reference.
Boundary markers: No specific delimiters or instructions to ignore embedded commands within user-provided data are mentioned in the documentation.
Capability inventory: The agent has significant capabilities, including arbitrary script execution (adobe_execute_script), file system writes/renders (adobe_render), and network-based asset installation (adobe_place_element).
Sanitization: There is no evidence of sanitization or filtering for the metadata ingested from project layers or markers before it is used in subsequent reasoning or script generation steps.

mira-after-effects