feat(amazon): Add stealth extraction for improved success rate (#31)

* docs: add Amazon stealth extraction design

Design for improving Amazon price extraction from ~10% to 50-60%
success rate using full stealth Playwright techniques:
- Browser identity rotation with fingerprint randomization
- Human-like behavior simulation (mouse, scroll, timing)
- Request strategy with graceful degradation

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* chore: add .worktrees to gitignore

Prepare for git worktree usage for feature development.

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* docs: add Amazon stealth implementation plan

8 TDD tasks covering:
- Browser identity profiles and rotation
- Human-like behavior simulation
- Stealth extraction with playwright-stealth
- Price service integration
- Monitoring and feature flags

Estimated: ~3 hours

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* deps: add playwright-stealth for Amazon extraction

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* feat(amazon): add BrowserIdentity dataclass and profile pool

Add BrowserIdentity dataclass to represent unique browser fingerprints
for stealth extraction, and IDENTITY_PROFILES list with 12 realistic
browser configurations (Chrome, Safari, Firefox, Edge across Mac,
Windows, and Linux).

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* feat(amazon): add IdentityManager with Redis persistence

Implements browser identity rotation and burn tracking:
- Rotates identities to avoid detection patterns
- Tracks request counts per identity in Redis
- Burns identities that trigger CAPTCHA for 24 hours
- Persists cookies per identity for session reuse
- Prefers lowest-usage identities for load balancing

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* feat(amazon): add human-like behavior simulation functions

Add behaviors.py with functions to simulate human browsing:
- human_delay(): Random delays with variance (returns seconds)
- generate_bezier_points(): Natural mouse movement paths
- human_mouse_move(): Move mouse along bezier curve
- human_scroll(): Human-like page scrolling
- handle_cookie_banner(): Dismiss Amazon cookie dialogs
- interact_like_human(): Combined human simulation sequence
- COOKIE_ACCEPT_SELECTORS: Amazon-specific cookie button selectors

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* feat(amazon): add stealth extractor with failure classification

- Add ExtractionResult dataclass for extraction outcomes
- Add AmazonFailureType enum (CAPTCHA, RATE_LIMITED, NO_PRICE_FOUND, NETWORK_ERROR)
- Add classify_failure() function to categorize extraction failures
- Add stealth_fetch_amazon() async function using playwright-stealth v2.0.1
- Add stealth_fetch_amazon_sync() synchronous wrapper
- Integrate with existing BrowserIdentity and IdentityManager
- Persist cookies between requests via identity_manager
- Update package __init__.py with new exports

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* feat(amazon): integrate stealth extraction into price service

- Add AMAZON_STEALTH_ENABLED feature flag (default: True)
- Add _get_identity_manager() singleton for lazy initialization
- Modify _fetch_amazon_price to use stealth when enabled
- Rename original logic to _fetch_amazon_price_legacy for fallback
- Mark identity as burned on CAPTCHA detection
- Fall back to legacy extraction when stealth unavailable
- Add 5 integration tests covering all code paths

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* feat(amazon): add stealth metrics logging and config flag

- Add AMAZON_STEALTH_ENABLED environment variable to config.py
  (defaults to true, can be disabled via env var)
- Add log_stealth_extraction() function to price_metrics.py for
  monitoring stealth extraction attempts with identity tracking
- Update price_service.py to import feature flag from Config
  instead of using hardcoded value

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

* docs: add Amazon stealth extraction documentation

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.5 <noreply@anthropic.com>

Author: ssk42

.gitignore

@@ -116,3 +116,6 @@ test_output.txt
 test_run.log
 coverage.xml
 .coverage
+
+# Git worktrees
+.worktrees/

CLAUDE.md

@@ -152,6 +152,14 @@ The application prevents gift receivers from seeing who claimed/purchased their
 - Production (Heroku): PostgreSQL via `DATABASE_URL` environment variable
 - Automatic postgres:// to postgresql:// URI conversion for Heroku compatibility
 
+#### Amazon Stealth Extraction
+The application uses stealth Playwright techniques for Amazon price extraction:
+- **Browser Identity Rotation**: 12 realistic browser profiles rotated every 10-20 requests
+- **Human-like Behavior**: Mouse movements, scrolling, natural delays
+- **Identity Burn Tracking**: Identities that trigger CAPTCHA are disabled for 24 hours
+- **Feature Flag**: Controlled via `AMAZON_STEALTH_ENABLED` environment variable
+- **Implementation**: `services/amazon_stealth/` module
+
 ### Testing Architecture
 - **Unit tests** (`tests/unit/`): Use Flask test client with temporary SQLite database
 - **Browser tests** (`tests/browser/`): Playwright-based end-to-end tests with live test server on port 5001

config.py

@@ -96,6 +96,9 @@ def get_ratelimit_storage_uri():
     LOG_LEVEL = os.getenv('LOG_LEVEL', 'INFO')
     LOG_FILE = os.getenv('LOG_FILE', 'wishlist.log')
 
+    # Amazon stealth extraction settings
+    AMAZON_STEALTH_ENABLED = os.environ.get('AMAZON_STEALTH_ENABLED', 'true').lower() == 'true'
+
     # Security Headers (configured in app.py)
     SECURITY_HEADERS = {
         'X-Content-Type-Options': 'nosniff',