The Agent Skills Directory

[SAFE]: The skill is designed for static text processing and metadata management. It explicitly disclaims any ability to generate media, synthesis audio, or connect to external APIs, which significantly limits its attack surface.
[COMMAND_EXECUTION]: Includes local validation scripts (scripts/asset_check.py and scripts/selftest.py) intended for use by the developer/creator. Analysis of the Python code confirms they only use standard libraries (argparse, json, pathlib) for structural integrity checks and do not contain dangerous calls such as os.system, subprocess, or eval.
[PROMPT_INJECTION]: The skill processes external, untrusted script data, representing a surface for indirect prompt injection. However, this is mitigated through a robust security architecture:
Ingestion points: Processes script files (screenplay.md) and index JSONL files.
Boundary markers: Implements a strict "Snapshot" contract where every referenced file must be declared in a sources block with a SHA256 hash. Data is categorized into distinct record types (Character, Look, Prop) with stable ID prefixes.
Capability inventory: Capabilities are limited to local text and JSONL generation. The skill does not possess network access or the ability to execute code based on the ingested content.
Sanitization: Uses the provided asset_check.py script to perform deep structural validation, ensuring all references are resolved within the declared snapshots and maintaining logical consistency in the asset graph.

short-drama-assets