The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill downloads an external codebase from an unverified GitHub repository (kimtaeyoon83/mcp-server-youtube-transcript) to provide core functionality.
[REMOTE_CODE_EXECUTION]: It executes Node.js code from the downloaded unverified repository at runtime using the node -e command.
[COMMAND_EXECUTION]: The instructions require the agent to perform high-privilege tasks within the /root/ directory, including cloning repositories and installing software dependencies.
[COMMAND_EXECUTION]: The workflow involves running arbitrary shell commands like npm install and npm run build on code fetched from an external source.
[DATA_EXFILTRATION]: The skill has the capability to send local files and transcripts to external messaging platforms (specifically Telegram) using the message command.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests and processes untrusted YouTube transcripts without adequate sanitization or boundary markers.
Ingestion points: The skill fetches transcript text from an external YouTube fetcher script.
Boundary markers: There are no delimiters or instructions to ignore potential commands embedded within the transcript content.
Capability inventory: The skill has access to shell command execution, file system modification in sensitive directories, and external network messaging.
Sanitization: No evidence of transcript content validation or escaping is present before the data is processed or summarized.

youtube-summarizer