The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: Automated security scan detected a blacklisted URL (https://url3.com) used as a placeholder in documentation examples within README.md.\n- [EXTERNAL_DOWNLOADS]: The skill retrieves content from user-provided URLs, including YouTube transcripts, PDFs, and browser extensions, as part of its core parsing functionality.\n- [COMMAND_EXECUTION]: The workflow logic involves executing several CLI tools (such as unzip, grep, pdftotext, and js-beautify) on files retrieved from remote sources to extract and process data.\n- [PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface by ingesting untrusted data from the web and interpolating it into prompts for large language models to perform entity extraction and summarization. Ingestion points: Workflows/ExtractArticle.md, Workflows/ExtractNewsletter.md, and Workflows/ExtractYoutube.md. Boundary markers: Prompts use delimiters like triple quotes and specific headers to isolate untrusted content. Capability inventory: Includes network access via curl, file system operations, and subprocess calls for data processing. Sanitization: The skill performs schema validation on LLM outputs using the ajv library but lacks active input sanitization for embedded instructions.

Parser