The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands to manage dependencies and process data. It runs pip install youtube-transcript-api to install required libraries and uses python -c to execute an embedded script for extracting and saving video transcripts to the local filesystem.\n- [EXTERNAL_DOWNLOADS]: The skill downloads code and data from external services. It pulls the youtube-transcript-api package from public registries and utilizes the Exa search tool to fetch metadata about YouTube videos from the web.\n- [PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it processes untrusted data from the web (via WebFetch and transcripts) without explicit boundary markers or sanitization, which could allow malicious instructions in the content to influence the agent's behavior.\n- [DATA_EXFILTRATION]: The skill requests access to sensitive local files, including the user's Claude memory file (~/.claude/memory/MEMORY.md) and a skill index file. While these are used for context, accessing such files exposes private information to the LLM's processing context.

triage-finding