The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill utilizes the WebFetch tool to retrieve taxonomy definitions from external URLs and research video ads on platforms including YouTube, Meta Ad Library, and iSpot.tv.
[COMMAND_EXECUTION]: The scripts/preflight.py script invokes the cp command via subprocess.run to copy video files from ~/Downloads to a local working directory. This operation is a documented workaround for macOS Transparency, Consent, and Control (TCC) restrictions.
[COMMAND_EXECUTION]: The skill performs environment validation by executing a setup check script for the watch skill dependency located at ~/.claude/plugins/cache/claude-video/watch/*/scripts/setup.py.
[CREDENTIALS_UNSAFE]: The preflight script accesses ~/.config/watch/.env to verify the configuration of API keys for the watch skill. The script only checks for key presence and identifies the service provider (Groq or OpenAI) without exposing the keys themselves.
[PROMPT_INJECTION]: The skill processes untrusted data from external websites, creating an indirect prompt injection surface. This is mitigated by the architecture, which uses isolated sub-agents for data extraction and includes a human-in-the-loop gate for reviewing all synthesized ticket drafts before external transmission. Ingestion points: taxonomy URLs and external search results. Boundary markers: absent. Capability inventory: file copy operations and Linear API interaction. Sanitization: handled via LLM synthesis and mandatory human review.

commission-video-format-skill