The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes external transcript files to generate titles, representing an attack surface for indirect instructions. Ingestion points: transcript files (SRT, VTT, JSON) read in scripts/chapters.py. Boundary markers: The agent is instructed in SKILL.md to wait for explicit user confirmation ('ship it, or change it?') before performing actions. Capability inventory: The skill has tool access to update_post which modifies YouTube content. Sanitization: Keyword extraction logic in scripts/chapters.py acts as a basic text filter.
[DYNAMIC_EXECUTION]: The scripts/chapters.py script uses sys.path.insert to dynamically include a directory for importing local dependencies (deadair) based on the relative file system structure.
[COMMAND_EXECUTION]: The documentation provides shell command templates for yt-dlp and python3, and the test suite (tests/test_chapters_ko.py) utilizes subprocess.run to execute the skill's Python scripts.
[DATA_EXFILTRATION]: The instructions direct the agent to read voice.md from a hidden folder in the user's home directory (~/.uplika/), which constitutes sensitive file path access to user configuration data.

yt-chapters