The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script chapters.py dynamically modifies the Python search path using sys.path.insert(0, ...) to include the directory ../yt-edit. It then imports the deadair module from this location. Loading code from computed relative paths outside the skill's own directory is a security risk as it allows for potential execution of unauthorized code if a malicious file exists in the targeted path.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted transcript data (SRT, VTT, or JSON files) to generate chapter titles. These titles are subsequently passed to the AI agent to be rewritten in a specific voice. This creates an attack surface where malicious instructions could be embedded in the transcript text to influence the agent's behavior during the retitling phase.
Ingestion points: The chapters.py script reads a user-supplied file path provided as a command-line argument (sys.argv[1]).
Boundary markers: There are no explicit boundary markers or instructions in the prompt to the agent to distinguish between the generated chapter content and potentially malicious instructions embedded within the transcript data.
Capability inventory: The skill has file system read access via os.path.exists and load(), and it outputs text to the standard output for agent consumption.
Sanitization: The transcript text is processed using regular expressions to extract keywords, but no security-focused sanitization or filtering is performed on the content itself.

yt-chapters