The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill dynamically loads and executes Python code from a local file named scene.py to define the film's visual and temporal logic (see scripts/render.py). It also uses dynamic imports in scripts/check_setup.py to verify the presence of required libraries.
[COMMAND_EXECUTION]: Multiple scripts (including scripts/build_narration.py, scripts/render.py, and scripts/mix_soundtrack.py) execute shell commands using subprocess.run to perform media encoding and analysis with FFmpeg. Notably, scripts/tts_say.py allows the execution of arbitrary shell commands via the --cmd parameter with shell=True, which creates a risk of command injection if the input string is manipulated by malicious data.
[EXTERNAL_DOWNLOADS]: The skill fetches resources from external services, including open-licensed fonts from Google's official GitHub repository (scripts/get_fonts.py), geographic boundary data from jsdelivr (scripts/make_map.py), and archival images from Wikimedia Commons (scripts/fetch_commons.py). These downloads are consistent with the skill's stated purpose and target well-known technology services and repositories.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user-provided files like script.json and photos.json to drive the film generation engine. This creates a potential surface for indirect prompt injection.
Ingestion points: Data enters the agent's context from local JSON project files and metadata retrieved from the Wikimedia Commons API.
Boundary markers: There are no explicit boundary markers or instructions provided to the agent to ignore potentially malicious embedded content within these data files.
Capability inventory: The skill possesses capabilities including executing shell commands (via FFmpeg and TTS tools), writing to the local filesystem, and making network requests for asset retrieval.
Sanitization: No evidence of sanitization or validation of the natural language content within the ingested JSON files was found.

cinematic-film