The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes several system binaries and CLI tools to perform its tasks:
yt (Fabric) for fetching YouTube transcripts.
llm CLI for communicating with Gemini models.
bun for running TypeScript utility scripts.
pdftotext and PyPDF2 (Python) for processing PDF documents.
unzip, diff, grep, and js-beautify for browser extension security analysis.
[EXTERNAL_DOWNLOADS]: The skill interacts with external domains to fetch content for parsing and use platform APIs:
Fetches web content, transcripts, and documents from user-provided URLs.
Interacts with Google's Gemini API at generativelanguage.googleapis.com (well-known service).
Downloads browser extension files from clients2.google.com (well-known service).
[PROMPT_INJECTION]: The skill is susceptible to Indirect Prompt Injection (Category 8) as it processes untrusted data from the web:
Ingestion points: External content is fetched in various extraction workflows (e.g., ExtractArticle.md, ExtractYoutube.md).
Boundary markers: The skill uses markdown code blocks, triple quotes ("""), and clear "ARTICLE X" delimiters in prompts to separate untrusted content from system instructions.
Capability inventory: The agent has access to shell commands and network requests, which could be targeted by embedded instructions in the fetched content.
Sanitization: Uses html2text to strip HTML tags, which provides a layer of sanitization but does not prevent text-based injection attacks.
[DATA_EXFILTRATION]: Performs automated network requests to a local notification service:
Executes curl POST requests to http://localhost:8888/notify to provide status updates to the user's local infrastructure.
[SAFE]: Automated scanner detections for malicious URLs and infected files are determined to be false positives based on context:
The URL https://channelwebsite.com is used exclusively as a placeholder within a documentation example in Workflows/ExtractYoutube.md.
The infected file alert for Workflows/ExtractYoutube.md likely triggered on the presence of shell commands (curl) and placeholder URLs within the markdown documentation.

Parser