The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill scans .env files to detect the application's base URL. Although it targets non-sensitive keys like APP_URL, environment files often contain sensitive credentials and secrets, and direct access to them is a security risk.
[COMMAND_EXECUTION]: Uses the shell to execute npx --yes @playwright/cli@latest, which downloads and runs the Playwright command-line tool.
[EXTERNAL_DOWNLOADS]: Fetches the axe-core accessibility engine from cdnjs.cloudflare.com and installs @playwright/cli from the npm registry. Both are well-known and established services.
[REMOTE_CODE_EXECUTION]: Utilizes Playwright's run-code and evaluate methods to inject and execute JavaScript (axe-core) within a browser session to perform the audit.
[PROMPT_INJECTION]: The skill processes untrusted data from the web, creating an indirect prompt injection surface.
Ingestion points: Retrieves WCAG guidelines and framework documentation from external URLs using firecrawl:firecrawl_scrape.
Boundary markers: None. The skill lacks explicit instructions or delimiters to isolate retrieved external data from the agent's internal logic.
Capability inventory: Includes shell command execution, file system discovery via Glob and Grep, and browser-based code execution.
Sanitization: None. Data from external scrapes is used directly in the context without filtering or validation.

audit-accessibility