Автономний app walker

Autonomous walker explores bounded portion of web application UI in managed tester VM. Records states, transitions, screenshots, console and network diagnostics, deduplicated findings. Useful for discovery; not proof every route or behavior tested.

Availability: walker data plane, reports, authenticated profiles, bounded mechanical/AI/hybrid/task modes, safety controls, scenario mining implemented. Walk can still miss behavior because of limits, dynamic UI, inaccessible labels, unavailable credentials, model/parser error. Review coverage before relying on result.

Start safely

Use non-production environment and dedicated test account. Obtain tester VM, start with default mechanical policy:


ensure_tester_vm(app="my-app", app_url="https://staging.example.com")

walk_app(
  app_name="my-app",
  app_url="https://staging.example.com",
  vm_name="<tester-vm>",
  max_states=40,
  max_depth=5,
  max_wall_seconds=600,
  destructive_allowed=false
)

walk_app returns immediately. Keep walk_id, poll get_app_walk(walk_id=...); call stop_app_walk if run should end early.

Exploration modes

Mode	Behavior
`mechanical`	Bounded DOM-order traversal; no model ranking
`ai`	Model ranks candidate actions, falls back when call cap reached
`hybrid`	Model ranking for first configured states, then mechanical traversal
`cot`	Task-directed search with decision cap and optional output schema

AI and hybrid modes fall back to mechanical when no supported model backend available. Task-directed mode fails instead. Model-ranked actions remain subject to same origin and destructive-action policy.

Safety controls

Destructive actions skipped by default.
destructive_allowed=true permits only guarded class; high-stakes labels remain blocked unless explicitly allowlisted.
Task-directed mode never accepts destructive_allowed=true.
Form filling avoids sensitive fields and blocks transactional submits.
Navigation origin-limited; task mode may add explicit bounded origin allowlist.
Authentication configuration contains secret variable names, not literal credentials. Scenario runner resolves values from owner keystore.

Allowlist operator authorization decision. Use narrow literal labels, review target environment, prefer test data that can be discarded.

Reading the result

Completed walk reports what actually observed:

visited UI states and transitions;
screenshots and interactive-element fingerprints;
console and meaningful network errors;
findings with severity, category, evidence, deduplication key; and
configured state, depth, action, time, model, safety limits.

completed means frontier ended or configured limit reached without internal error. Does not mean application passed. Review blocker and major findings, skipped actions, authentication status, coverage totals.

Authenticated walks

Prefer encrypted browser profile when available. Legacy username/password flow accepts only keystore variable names. If sign-in fails or expected post-login content absent, run records authentication failure; do not interpret walk of login page as coverage of product.

Turn discovery into regression tests

After reviewing finding save deterministic app scenario for affected flow. Mined scenario can be low-confidence unless verification checkpoint passed, inspect steps before adopting as release gate.