The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the installation of an external dependency from an unverified GitHub repository.
Evidence: Instructions in SKILL.md and package.json to git clone https://github.com/kimtaeyoon83/mcp-server-youtube-transcript.git.
[REMOTE_CODE_EXECUTION]: The skill performs remote code execution by downloading and building an external package at runtime.
Evidence: git clone ... && npm install && npm run build commands provided in the installation workflow and package.json.
[COMMAND_EXECUTION]: The skill uses the shell to execute Node.js commands and platform-specific messaging tools.
Evidence: Usage of node --input-type=module -e to execute inline JavaScript and message --action send to interact with Telegram.
[DYNAMIC_EXECUTION]: The skill generates and executes JavaScript code at runtime by interpolating video IDs into a shell command.
Evidence: The inline script in SKILL.md that imports getSubtitles and processes output using node -e.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from YouTube transcripts which could contain malicious instructions designed to influence the agent's summary or behavior.
Ingestion points: Transcript data fetched via the mcp-server-youtube-transcript tool.
Boundary markers: None identified in the prompt templates to delimit external transcript content from agent instructions.
Capability inventory: File system writes to /root/, shell command execution via node, and network access via Telegram integration.
Sanitization: No evidence of sanitization for the transcript text before it is processed by the LLM.
[DATA_EXFILTRATION]: The skill is configured to send transcript files to external platforms using environment-stored credentials.
Evidence: SKILL-OC.md contains instructions to use curl to send documents to the Telegram Bot API using ${TELEGRAM_BOT_TOKEN}.

youtube-summarizer