The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires cloning a full codebase from a third-party repository (github.com/ok-helloworld/vibe-pentest or gitee.com/ok-helloworld/vibe-pentest) which is not an officially trusted organization or well-known service vendor.
[COMMAND_EXECUTION]: The workflow relies heavily on subprocess.run and subprocess.Popen to execute multiple Python scripts (fingerprint.py, admin_scanner.py, run_katana.py, prepare_agent_findings.py, generate_report.py) downloaded from the external repository.
[DYNAMIC_EXECUTION]: The skill uses Playwright's page.evaluate() to execute arbitrary JavaScript within a browser context for session extraction and authorization header capture.
[INDIRECT_PROMPT_INJECTION]: The skill is highly susceptible to indirect prompt injection because its core function is to ingest and process untrusted data from external web applications.
Ingestion points: Data enters the system via the Katana crawler, which saves results to workspace/crawled_anonymous.jsonl, and through the multi-agent findings stored in workspace/findings/*.json.
Boundary markers: None identified; the skill directly processes crawler output to map attack surfaces.
Capability inventory: The skill has broad capabilities including file system writes (workspace/), network requests via httpx and requests, and browser automation via playwright.
Sanitization: There is no evidence of sanitization or escaping of the content discovered by the crawler before it is passed to the specialized AI agents for vulnerability analysis.

vibe-pentest-ai-security-testing