The Agent Skills Directory

[Indirect Prompt Injection] (LOW): The skill ingests data from external websites, creating a potential vector for indirect prompt injection where malicious instructions embedded in a web page could influence the agent's downstream tasks.
Ingestion points: Web content is fetched from arbitrary URLs in scripts/scrape-and-convert.ts using Playwright.
Boundary markers: The skill formats output with URL headers but does not explicitly instruct the agent to ignore directions found within the scraped content.
Capability inventory: The skill uses Bash to run pnpm tsx, can write to the local filesystem (docs/web-captures/), and has network access via the headless browser.
Sanitization: The scraping process removes script, style, and noscript tags to prevent browser-side code execution but does not filter the text for natural language prompt injection patterns.
[Unverifiable Dependencies & Remote Code Execution] (SAFE): The skill installs Playwright and browser binaries from Microsoft repositories, which is considered a trusted source.
Evidence: SKILL.md instructions and package.json dependencies point to standard, reputable packages and official installation methods.

web-to-markdown