Discovery Session Best Practices

Discovery sessions are the foundation for automated flow generation. Well-planned sessions capture high-quality visual contexts that translate into reliable test flows.

Planning Your Discovery Strategy

Defining Discovery Scope

Balance exploration breadth against credit consumption and result quality:

Scope	Best For	Credit Impact
Targeted	Specific feature areas	Low (20-50 steps)
Moderate	Product section or user journey	Medium (50-100 steps)
Broad	Full application mapping	High (100+ steps)

When to use targeted discovery:

• Testing a new feature in isolation
• Re-exploring after UI changes
• Building flows for specific user journeys

When to use broad discovery:

• Initial application onboarding
• Quarterly coverage audits
• Major redesign verification

tip

Start with targeted discoveries for immediate needs. Run broad discoveries periodically to maintain comprehensive coverage.

Pre-Discovery Preparation

Prepare before starting to maximize session value:

1. Set up browser profiles

For authenticated areas, create a browser profile with active sessions:

• Log in manually using the profile
• Save cookies and session state
• Select this profile when starting discovery

See Browser Profiles for setup instructions.

2. Identify key user journeys

List the flows you want to generate:

• Core authentication paths
• Primary conversion funnels
• Frequently used features

3. Plan exclusion patterns

Identify URLs to skip:

• Admin panels and internal tools
• Third-party embedded content
• Logout and session-ending pages
• Marketing or promotional popups

4. Write custom instructions

Guide the AI agent's exploration focus:

Good: "Focus on the checkout flow and payment options.
       Skip promotional banners and newsletter popups."

Avoid: "Explore everything" (too vague)

Optimizing Visual Context Capture

Understanding Context Types

Discovery captures three types of visual contexts:

Type	Description	Flow Generation Value
Page	Full page views	Primary test flow sources
Component	Reusable UI elements	Shared validation patterns
State	Application states (error, success, loading)	Edge case and assertion sources

Maximize capture diversity:

• Run discoveries at different times (morning vs. evening traffic)
• Use different user accounts (new vs. returning user)
• Trigger various application states (empty cart, full cart, errors)

Maximizing Context Quality

Capture dynamic content:

• Allow time for lazy-loaded elements
• Interact with expandable sections
• Trigger form validation messages

Capture error states:

• Submit invalid data intentionally
• Access restricted pages
• Test with expired sessions

Handle forms thoughtfully:

• Include form interactions in custom instructions
• Capture both empty and filled states
• Trigger and capture validation errors

note

Visual contexts are deduplicated by (URL pattern, context type). Running multiple discoveries of the same pages won't create duplicates—it updates existing contexts with fresher data.

Discovery Session Configuration

URL Filtering Best Practices

Use exclusion patterns to focus exploration and save credits:

Common exclusion patterns:

# Administrative areas
https://example.com/admin/**
https://example.com/internal/**

# Authentication endpoints that end sessions
https://example.com/logout
https://example.com/signout

# External services
https://docs.example.com/**
https://help.example.com/**

# Marketing and promotional content
https://example.com/promo/**
https://example.com/campaign/**

Pattern syntax:

• ** matches any path segments
• * matches within a single segment
• Exact URLs match exactly

Project-level vs. session-level exclusions:

• Project-level: Persistent exclusions for areas you never want to explore
• Session-level: Temporary exclusions for specific discovery goals

Custom Instructions

Write concise, outcome-focused instructions:

Effective examples:

"Prioritize the account settings and profile editing flows.
Capture error states when saving invalid data."

"Focus on the product catalog and filtering options.
Skip the blog and support sections."

"Explore the checkout process thoroughly.
Include guest checkout and registered user paths."

Instruction tips:

• Keep under 200 characters for clarity
• Specify what to include AND what to skip
• Mention specific features or pages by name
• Request specific states to capture (errors, empty states)

Browser Profile Selection

Match profiles to discovery goals:

Goal	Profile Choice
Public pages	Default (temporary)
Authenticated user flows	User account profile
Admin functionality	Admin account profile
Different user roles	Create profile per role

caution

Using the wrong profile wastes credits. If discovery can't access authenticated areas, it explores only public pages and misses valuable contexts.

Monitoring and Managing Sessions

Real-Time Monitoring

Watch these metrics during discovery:

Metric	What It Tells You	Action If Concerning
Steps	Credit consumption rate	Stop if progressing too slowly
Visual Contexts	Valuable captures	Good if growing steadily
Agent Focus	Current exploration area	Verify it matches your goals
Last Action Result	Success/failure status	Stop if repeatedly failing

Signs to stop early:

• Agent stuck in a loop (visiting same pages repeatedly)
• Authentication expired (redirecting to login)
• High step count with few visual contexts
• Agent exploring excluded or irrelevant areas

Interpreting Session Results

After discovery completes, evaluate results:

Good session indicators:

• Visual contexts cover target areas
• Mix of page, component, and state types
• Agent reasoning shows logical exploration
• Reasonable step-to-context ratio

Poor session indicators:

• Few visual contexts despite many steps
• Contexts only from login or error pages
• Agent stuck on single page or section
• Missing expected application areas

Static vs. Dynamic Analysis

Static Analysis Capabilities

Discovery begins with static analysis:

• robots.txt parsing: Respects crawl directives
• XML sitemap discovery: Uses sitemap URLs for initial exploration
• Template detection: Identifies repeated components (headers, footers, nav)

Leverage static analysis:

• Ensure your sitemap is current and complete
• Use robots.txt to guide (not block) the agent
• Template detection reduces redundant captures

Dynamic AI-Powered Crawling

After static analysis, AI explores dynamically:

How the agent explores:

1. Analyzes visible elements and interactions
2. Clicks links, buttons, and interactive elements
3. Fills forms with reasonable test data
4. Captures visual contexts at significant states
5. Follows navigation patterns like a real user

Single Page Application (SPA) considerations:

• Agent handles client-side routing
• May need longer timeouts for initial load
• Custom instructions can guide through app sections

JavaScript-heavy applications:

• Agent waits for dynamic content to load
• Complex interactions may need custom instructions
• Consider targeted discoveries for complex features

Post-Discovery Workflow

Visual Context Review

Before generating flows, review captured contexts:

Quality checklist:

• Contexts cover intended application areas
• Mix of page types (not just homepage)
• Includes interactive states (forms filled, menus open)
• Error states captured for edge case flows
• No irrelevant or duplicate contexts

Organizing contexts:

• Note which contexts map to planned flows
• Identify gaps requiring additional discovery
• Mark high-priority contexts for immediate flow generation

Flow Generation Strategy

When to generate flows immediately:

• Contexts cover a complete user journey
• Testing deadline requires quick turnaround
• Baseline coverage needed urgently

When to delay flow generation:

• Missing key application areas
• Planning additional discovery sessions
• Need team review of contexts first

Generation approach:

• Generate flows from ALL contexts in the project
• Review and refine generated flows (1 credit per refinement)
• Prioritize flows for critical user journeys

Common Discovery Pitfalls

Pitfall 1: Scope Too Broad

Problem: Discovery explores entire site, consuming hundreds of credits with unfocused results.

Solution: Use targeted scope with URL exclusions. Run multiple focused discoveries instead of one massive session.

Pitfall 2: Missing Authentication

Problem: Discovery captures only public pages; authenticated areas produce no contexts.

Solution: Create and select a browser profile with active login session before starting discovery.

Pitfall 3: No URL Exclusions

Problem: Agent wastes steps on admin panels, logout pages, or external links.

Solution: Define exclusion patterns for irrelevant areas. Add project-level exclusions for persistent rules.

Pitfall 4: Single Role Discovery

Problem: Discovery with one user role misses features available to other roles.

Solution: Run separate discoveries with different browser profiles for each user role (customer, admin, guest).

Pitfall 5: Ignoring Session Metrics

Problem: Session runs to completion despite poor progress, wasting credits.

Solution: Monitor metrics in real-time. Stop early if step-to-context ratio is poor or agent is stuck.

Discovery Session Metrics

Use these benchmarks to evaluate session quality:

Metric	Good	Concerning	Action
Visual Contexts / Steps	> 0.3	< 0.1	Refine scope, check auth
Steps to first context	< 10	> 25	Check start URL, exclusions
Context type diversity	All 3 types	Only 1 type	Adjust custom instructions
Coverage of target areas	> 80%	< 50%	Run additional targeted discovery

Next Steps

• Discovery Sessions User Guide - Full feature reference
• Flow Design Best Practices - Design effective test flows
• Credit Optimization - Minimize discovery costs
• Browser Profiles - Set up authenticated discovery