Configuration

AutoPOM-Agent behavior is controlled by CrawlConfig.

Core options

Setting	Purpose	Default
base_url	Entry URL for crawl	required
max_depth	Maximum recursive depth	3
max_pages	Maximum modeled pages	80
max_actions_per_page	Action budget per page	12
same_origin_only	Restrict to same domain	true
denied_domains	External domains denylist	social domains
output_dir	Output root for artifacts	output
pom_language	Generated POM language (java, javascript, typescript)	java
locator_storage	Selector storage strategy (inline, external)	inline
browser_adapter	Browser backend (mock, playwright)	mock
playwright_headless	Run Playwright headless (true) or headed (false)	true
cdp_url	Connect to existing browser via CDP URL	None
chrome_profile	Use local Chrome profile for sessions	false
interactive_pause	Wait for user input before capturing	false

Credentials

Use .env variables for gated flows:

AUTOPOM_USERNAME=...
AUTOPOM_PASSWORD=...

CLI equivalent

PYTHONPATH=src python3 -m autopom.cli.main \
  --base-url "https://example.com" \
  --output-dir "output" \
  --max-depth 3 \
  --max-pages 20 \
  --pom-language "javascript" \
  --browser-adapter "playwright" \
  --interactive

Advanced Browser Flags

Interactive Mode (--interactive)

Launches a headed browser and pauses execution. This is the recommended way to handle complex authentication (MFA, Captchas) or to navigate to a specific application state before triggering the crawl.

Existing Browser (--capture or -c)

Attaches to a browser already running with the --remote-debugging-port=9222 flag.

• Usage: --capture http://localhost:9222
• Default if no URL provided: http://localhost:9222

Chrome Profile (--chrome-profile)

Launches a new instance using your default Google Chrome profile. This inherits all your active sessions, cookies, and stored credentials. Note: All existing Chrome windows must be closed before using this flag to avoid profile lock errors.

Generation Options

Locator Storage (--locator-storage)

• inline (default): Locators are defined directly within the Page Object methods or fields.
• external: Locators are stored in a separate metadata file (coming soon for all languages), allowing for easier updates without modifying code.

Language aliases

Use bash run.sh for an interactive flow with enterprise-style prompts:

• URL, language, adapter, depth/pages
• transparent package/browser setup
• final execution preview before run

Language aliases

• js normalizes to javascript.
• ts normalizes to typescript.
• Unsupported values raise a validation error before crawl execution.

Browser adapter options

• mock: deterministic adapter for local testing and CI-safe runs.
• playwright: real browser crawl against live web applications.
• CLI shortcut: --headed turns off headless mode for Playwright runs.

Reporting outputs

Every run writes:

• reports/crawl_summary.md for crawl quality and selector confidence.
• reports/execution_summary.md for stakeholder-friendly run summary.
• reports/execution_summary.json for downstream analytics/dashboards.

Recommended policy baseline

• Keep max_depth between 2 and 4.
• Keep same_origin_only enabled.
• Add explicit denylist for external auth/marketing domains.
• Enforce crawl timeout budget in CI runners.