The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/breakdown_tool.py script executes system binaries (ffmpeg and ffprobe) using the subprocess module to perform video and audio analysis. The implementation follows security best practices by passing arguments as lists and avoiding shell=True, which mitigates command injection risks. The tool validates numerical inputs such as thresholds and crop coordinates before interpolation into filter strings.
[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection because it processes untrusted external data, specifically video transcripts (SRT, WebVTT, JSON) and drama scripts. Ingestion points: Transcripts and script files are parsed by breakdown_tool.py and the extracted content is provided to the agent via measured JSON files. Boundary markers: The instructions in references/breakdown-dimensions.md provide a structured reporting method (Direct Observation, Interpretation, Unknown) which helps the agent distinguish between raw data and subjective analysis. Capability inventory: The skill can execute media processing commands and write analysis files (JSON, JPG, MP4) to local directories. Sanitization: The Python script uses regular expressions to strip markup and non-spoken content from transcripts. No active exploitation was detected.

short-drama-breakdown