The Agent Skills Directory

[DYNAMIC_EXECUTION]: The helper scripts in the scripts/ directory (e.g., auth-analysis.sh, injection-probes.sh, logic-tester.sh) use bash -lc to execute assembled command strings. This is a common pattern for CLI-based utility skills. The scripts utilize printf %q in Bash to safely escape user-supplied inputs and target URLs before they are included in command strings, mitigating the risk of command injection.
[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill relies on and references several well-known external security tools including Playwright, OWASP ZAP, sqlmap, feroxbuster, ffuf, and jwt-tool. The instructions neutrally guide the user or agent to use these established third-party services which are standard in the cybersecurity industry.
[INDIRECT_PROMPT_INJECTION]: The skill's crawling component, playwright-crawl.mjs, ingests data from external web applications, such as page titles and form metadata. While this represents a surface for indirect prompt injection—where a malicious site could provide crafted content to influence the agent's logic—the risk is managed by the skill's specific focus on technical data extraction and its structured reporting format.
[DYNAMIC_EXECUTION]: The playwright-crawl.mjs script uses Playwright's page.$$eval and page.goto to execute JavaScript in a browser context for the purpose of identifying forms and links. This is standard behavior for a web crawler and is contained within the browser session targeting the specified pentesting URL.

pentesting-web-apps