PentesterFlow/agent — reverse-engineered prompt

Build me a terminal app for authorized security testing that feels like an AI assistant for pentesters. I want to open it, choose a local or compatible AI model, set a target URL, then type a plain English goal like testing an API for broken access control. The app should plan the work, pick a relevant testing playbook, ask me before running anything risky, run visible shell or HTTP checks, show the results clearly, and only save a finding after it has actually verified the issue.

It should support local model tools like Ollama or LM Studio, plus compatible hosted APIs. Include simple commands for setting the target, changing model, listing skills, resuming a session, enabling browser capture, and turning on a lab only mode where approvals can be skipped. Findings should be written as clean Markdown with evidence, impact, proof of concept, and remediation.

Make it installable on Mac, Linux, and Windows, and keep the workflow transparent and safe for scoped testing only.