sequenceDiagram
participant U as User (CLI)
participant M as interface/main.py
participant SS as scan_setup.py
participant Run as core/runner.py
participant SM as runtime/session_manager.py
participant Cai as Caido proxy (sandbox)
participant Root as Root Agent
participant Child as Child Agent(s)
participant Sbx as Docker sandbox (shared)
participant Rep as report/state.py + writer.py
U->>M: strix --target ./app-directory
M->>SS: prepare_run(args) -> run_name, local_sources
M->>Run: run_strix_scan(scan_config, ...)
Run->>SM: create_or_reuse(scan_id, image, local_sources)
SM->>Sbx: create container (bind mount, NET_ADMIN/RAW, proxy env)
SM->>Cai: bootstrap_caido() -> guest login, temp project
Run->>Root: build_strix_agent(is_root=True) + run_agent_loop()
Root->>Root: view_agent_graph, decompose target
Root->>Child: create_agent(name, task, skills)
activate Child
Child->>Sbx: exec_command (nmap, ffuf, sqlmap, agent-browser, python3)
Sbx-->>Cai: outbound HTTP intercepted (http_proxy env)
Child->>Rep: create_vulnerability_report(...)
Rep->>Rep: dedupe.check_duplicate() (LLM judge)
Rep-->>Child: accepted -> vuln-NNNN
Child->>Root: agent_finish() -> completion report (mailbox)
deactivate Child
Root->>Root: wait_for_agents() wakes, aggregates
Root->>Rep: finish_scan(executive_summary, methodology, ...)
Rep->>Rep: write_run_record, write_executive_report,\nwrite_vulnerabilities, write_sarif_report
Run->>SM: cleanup(scan_id) -> teardown sandbox
Run-->>U: penetration_test_report.md, vulnerabilities/*.md,\nvulnerabilities.json, findings.sarif (exit code 2 if findings)