The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill provides tools for dynamic code execution within the browser environment to handle complex interactions.
Evidence: The playwright_evaluate tool allows the agent to execute arbitrary JavaScript code within the page context.
Evidence: The browser_run_code tool enables the execution of multi-step Playwright scripts for complex workflows.
[PROMPT_INJECTION]: The skill is exposed to indirect prompt injection risks as it ingests content from external, untrusted web pages.
Ingestion points: Page content and network responses are retrieved via Playwright in SKILL.md and reference/data-extraction.md.
Capability inventory: The skill has the power to navigate, click, fill forms, and execute scripts based on interpreted content.
Sanitization/Markers: The skill references a prompt-injection classifier and a web fetch safety protocol in _common/WEB_FETCH_SAFETY.md.
[EXTERNAL_DOWNLOADS]: The skill documentation references the installation of external Node.js packages for automation and anti-detection.
Packages: playwright, playwright-extra, and puppeteer-extra-plugin-stealth are listed in the stealth mode reference.
[DATA_EXFILTRATION]: The skill is designed to collect data from web pages and save it to the local filesystem, which could be used to harvest sensitive information.
Destination: Extracted data is saved in structured formats (JSON/CSV) to the .vector/ directory.
Safeguards: The skill mandates the use of environment variables for credentials and respects robots.txt/ToS signals.

vector