The Agent Skills Directory

[PROMPT_INJECTION]: The skill facilitates an indirect prompt injection surface by instructing the agent to fetch and process content from arbitrary, user-supplied URLs.
Ingestion points: Untrusted data enters the agent context via page.goto(url) and document.body.innerText extraction in references/playwright-analysis.md.
Boundary markers: Absent; there are no instructions to use delimiters or ignore instructions found in external content.
Capability inventory: The environment allows package installation (npm install), browser engine downloads (npx playwright install), and local script execution (node analyze.js) with file system writes (fs.writeFileSync).
Sanitization: Absent; the skill does not define methods for validating or escaping external web content.
[COMMAND_EXECUTION]: The skill defines a sequence of shell commands to set up an analysis environment and execute a browser-based auditing script.
Evidence: references/playwright-analysis.md includes commands for npm install, npx playwright install, and script execution via node.
[EXTERNAL_DOWNLOADS]: The analysis workflow triggers the download of the playwright library and the Chromium browser from public package registries.
Evidence: Commands in references/playwright-analysis.md automate the retrieval of these dependencies from well-known registries.
[REMOTE_CODE_EXECUTION]: The skill employs a pattern of generating and executing a local script to interact with and process data from the internet.
Evidence: The workflow defined in references/playwright-analysis.md involves writing a functional JavaScript file to the disk and executing it in the local environment.

ai-seo