The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill audits external websites and processes their content, which could theoretically contain malicious instructions targeting the AI agent.
Ingestion points: scripts/observe-runtime.mjs extracts data from the target website's DOM via page.evaluate(), including text content, aria-labels, button names, and web storage data.
Boundary markers: Extracted content is returned to the agent in structured reports, but without explicit security delimiters to prevent the agent from following instructions embedded in the site's text.
Capability inventory: The skill can execute local scripts, launch a headless browser, write screenshot files, and is intended to modify project source code (CSS and HTML).
Sanitization: The tool performs basic sanitization by truncating extracted strings and redacting likely secrets (tokens) from storage data.
[COMMAND_EXECUTION]: The skill's primary functionality involves executing a local Node.js script that launches a browser process.
Evidence: SKILL.md instructs the agent to run observe-runtime.mjs, which in turn launches Chromium binaries found on the local system.
[DYNAMIC_EXECUTION]: The skill dynamically loads the Playwright library from the host environment at runtime.
Evidence: scripts/lib/browser.mjs uses createRequire and dynamic import() to resolve and load playwright or playwright-core from local node_modules or global paths.

rtl-readiness