The Agent Skills Directory

[COMMAND_EXECUTION]: The Python script scripts/generate_bundle.py uses string templates to generate bash scripts (.ffmpeg.sh). The {audio} and {srt_path} variables are interpolated directly into the template inside double quotes. Because bash evaluates commands inside double quotes (e.g., using $(...) or backticks), a maliciously crafted file path could lead to arbitrary command execution when the generated script is run by the user.
[DYNAMIC_EXECUTION]: The skill dynamically generates executable shell scripts at runtime. The scripts/generate_bundle.py script specifically uses chmod(0o755) on these generated files to enable execution. While intended for automation, this pattern creates a risk if the generated content is influenced by untrusted input.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external transcript files (Whisper JSON or SRT), which act as untrusted data ingestion points.
Ingestion points: scripts/generate_bundle.py (via --chapters and --clips arguments) and scripts/score_segments.py (which processes the main transcript content).
Boundary markers: No specific boundary markers or 'ignore' instructions are used to distinguish transcript data from instructions during processing.
Capability inventory: The skill performs file system writes and generates executable shell scripts.
Sanitization: There is no evidence of shell-safe escaping for the data interpolated into generated shell scripts.

hebrew-podcast-postproduction