Next.js (App Router) + Mistica web app for AgentHound. The threat-model wizard uploads an architecture YAML, calls the backend analysis API, and renders the capability graph, findings, and risk metrics.
For the full-stack setup (backend + frontend together), see the root README.
- Node.js 20+
- pnpm (the repo pins
pnpm@11) - A running backend (default
http://localhost:8000; see the root README)
pnpm install # install dependencies
pnpm dev # start the dev server on http://localhost:3000
pnpm build # production build
pnpm start # serve the production build
pnpm test:ts # type-check (tsc --noEmit)
pnpm lint # eslint
pnpm test # unit (Node) and component (jsdom) tests with Vitest
pnpm test:visual # Playwright visual regression tests against a production buildThe visual baselines (tests/visual/**/*-chromium-darwin.png) were generated on
macOS, so for now pnpm test:visual only passes there. On Linux or Windows,
generate local baselines first with pnpm test:visual:update and do not commit them.
The visual run builds into .next-visual/, so it leaves the regular .next/ alone.
The backend base URL defaults to http://localhost:8000. Override it with an
environment variable (e.g. in .env.local):
NEXT_PUBLIC_AGENTHOUND_API_URL=http://localhost:8000The API layer lives in src/lib/api/:
client.ts—fetchwrapper; resolves the base URL and maps the backend error envelope to anApiError.agentHound.ts— endpoint calls:createAnalysis(parse only),analyze(run the rule engine, optionally over the edited graph),getGraph,getFindings,getFinding(one finding plus its path subgraph, drawn alone from the graph toolbar),getControlRecommendations,simulateControls.types.ts— raw backend (snake_case) response shapes.adapters.ts— maps backend shapes to the camelCase UI types. Because the backend graph carries no coordinates,toGraphNodesassigns nodex/ywith a swimlane layout (one lane per node type); attack overlays are derived from critical/high finding paths.toSolutionResultmaps the recommend + simulate responses into the Proposed Solution (before/after metrics, remediation timeline, and the mitigated graph withblocked/mitigatededges and applied-control nodes).
The wizard's Inference step is backed by the real parsed graph: createAnalysis parses the YAML, the step shows the inferred per-agent capabilities for review/correction, and on continuing to Results analyze runs the rule engine over the (possibly edited) graph — so the findings, scores, and counts follow the user's edits. The Proposed Solution tab is wired to the Phase 2 endpoints.
The Results step splits findings into two sections keyed on each finding's findingClass (attack / hygiene, mapped from the backend finding_class): exploitable attack paths are shown first, hygiene and compliance signals second, so the hygiene volume does not bury the critical paths.