The Agent Skills Directory

[COMMAND_EXECUTION]: The skill makes extensive use of system commands to process media files. While expected for a media skill, these represent a significant capability surface.
Evidence: scripts/_mediaskills_common.py contains a run wrapper that executes commands using subprocess.run. This is used throughout the skill to call ffmpeg and ffprobe for frame extraction and media probing.
Evidence: scripts/compile_report.py executes tesseract for OCR analysis of extracted frames.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted media content and extracts text via OCR or agent vision, creating a surface for indirect prompt injection.
Ingestion points: Media files provided via the --input argument are processed into frames in scripts/extract_interval_frames.py and scripts/compile_report.py. Agent-generated analysis is ingested in scripts/merge_analysis.py.
Boundary markers: The skill does not implement explicit boundary markers or "ignore embedded instructions" warnings when presenting extracted text or frame descriptions to the agent.
Capability inventory: The skill possesses file-write capabilities (writing to .mediaskills/generated/) and command execution (FFmpeg/Tesseract).
Sanitization: scripts/_analysis_lib.py includes a coerce_frame_analysis function that cleans up types and lengths, but does not sanitize the actual text content for malicious instructions.
[EXTERNAL_DOWNLOADS]: The test suite includes a mechanism to synchronize shared libraries from a repository root.
Evidence: tests/test_vision_analysis.py contains a sync_common fixture that executes sync_shared_libs.py from the repository root using the current Python interpreter.

vision-analysis