The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's operational setup in references/playwright-analysis.md requires the agent to download the playwright package from the NPM registry and fetch browser binaries using npx playwright install chromium during the session.
[COMMAND_EXECUTION]: The skill instructs the agent to perform multiple shell commands to set up the analysis environment and run the scraping script, including mkdir, npm install, and node analyze.js.
[REMOTE_CODE_EXECUTION]: A full Node.js script is provided in references/playwright-analysis.md to be saved and executed. This script utilizes the fs module to write files and the playwright library to navigate to and interact with arbitrary URLs provided by the user.
[DATA_EXFILTRATION]: The provided analyze.js script collects extensive information from target websites, including rendered HTML, visible text, console logs, and full-page screenshots. This data is saved to the local file system and subsequently read by the agent, posing a risk of internal data exposure if the tool is used to target internal network addresses or sensitive files.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests and processes content from external, untrusted web pages without sufficient sanitization or boundary markers.
Ingestion points: Rendered text and screenshots retrieved from external URLs by the script in references/playwright-analysis.md.
Boundary markers: None; the agent is directed to review the collected content directly to inform its output.
Capability inventory: Subprocess execution, file system writes (fs.writeFileSync), and network navigation via Playwright.
Sanitization: None; the raw rendered content of the page is provided to the agent context.

competitors