S0 through S15: every pipeline stage, the real files that implement it, and what each key function does in plain English. See docs/design/data-flow.md for the full narrative.
The command line reads what you typed and checks Docker is ready before anything else happens.
strix/interface/main.py
main() — Top-level dispatch: routes to `strix view`, `strix auth`, or a scan._bootstrap_scan() — Runs preflight checks, then hands off to the CLI or TUI.strix/interface/cli_args.py
parse_arguments() — Reads every --flag you can pass to strix.Strix decides what to call this run and figures out exactly what to mount into the sandbox.
strix/interface/scan_setup.py
prepare_run() — Generates the run name, resolves local source paths, computes diff-scope for CI.Strix opens a blank notebook for this run's findings before anything starts.
strix/report/state.py
ReportState.__init__() — Creates the in-memory findings tracker for one run.get_global_report_state() — Where every other part of Strix looks up the current run's notebook.A disposable container is started and loaded with the real hacking tools, plus a proxy that watches all its traffic.
strix/runtime/session_manager.py
create_or_reuse() — Builds the sandbox manifest and starts (or reuses) the container.strix/runtime/docker_client.py
StrixDockerSandboxClient._create_container() — The actual Docker container creation, with raw-socket capabilities for nmap.strix/runtime/caido_bootstrap.py
bootstrap_caido() — Logs into the in-container proxy sidecar so it's ready to intercept traffic.Strix builds its 'team lead' AI agent: gives it its tools and its job description.
strix/agents/factory.py
build_strix_agent() — Assembles the tool list and the lifecycle stop-condition for one agent.strix/agents/prompt.py
render_system_prompt() — Turns the agent's role + relevant skill files into its actual instructions.The lead agent starts working — reading, thinking, and calling tools, one turn at a time.
strix/core/execution.py
run_agent_loop() — Drives one agent's turns until it calls a lifecycle tool or hits a limit.The lead reads the brief and decides which kinds of testing are worth doing.
strix/agents/prompts/system_prompt.jinja
<root_agent_directive> — The instructions telling the root agent it must delegate, never test directly.Every time the lead decides on a specialist, one gets hired and starts working immediately, in parallel.
strix/tools/agents_graph/tools.py
create_agent() — The tool call the root agent uses to hire a specialist.strix/core/execution.py
spawn_child_agent() — Builds the child, registers it, and launches its own turn loop as a background task.Specialists actually run the tools — port scans, injection attempts, browser automation — for real, inside the shared sandbox.
strix/tools/proxy/tools.py
repeat_request() — Replays and modifies a captured HTTP request — the 'Repeater' feature.strix/tools/shell/
exec_command — SDK-native — runs any shell command inside the sandbox.Before filing anything, the specialist has to actually prove the bug is real.
strix/tools/reporting/tool.py
_do_create() — Validates every required field before a finding can even be considered.The finding gets scored, checked against everything already found, and — if new — saved.
strix/report/dedupe.py
check_duplicate() — An AI judge asks: is this the same root cause on the same asset as something already found?strix/report/state.py
add_vulnerability_report() — Assigns the next vuln-NNNN id and stores the full finding.The specialist wraps up and mails a summary back to whoever hired it.
strix/tools/agents_graph/tools.py
agent_finish() — Renders a completion report and sends it to the parent's inbox.The lead wakes up, reads what the specialist found, and decides whether more specialists are needed.
strix/tools/agents_graph/tools.py
wait_for_agents() — The blocking call the lead uses to wait for a specialist's report.The lead decides testing is done and writes the executive summary.
strix/tools/finish/tool.py
finish_scan() — Root-only — validates the summary fields and ends the scan.Everything found gets written to disk in a form a human or a CI pipeline can actually read.
strix/report/writer.py
write_executive_report() — Writes the narrative report.write_vulnerabilities() — Writes one Markdown file per finding, plus CSV/JSON indexes.strix/report/sarif.py
write_sarif_report() — Writes the CI/CD-ready SARIF file.The disposable container and everything in it is deleted — nothing about the target's environment is left behind.
strix/runtime/session_manager.py
cleanup() — Closes the Caido client and deletes the Docker container.