Chaos Runs

Autonomous AI-powered exploratory testing.

Chaos Runs

Chaos runs launch autonomous AI agents that explore your application, discover features, and test them from multiple angles. Unlike scripted test runs, chaos runs require no pre-defined workflows - the agents figure out what to test on their own.

Starting a Chaos Run

  1. Go to Runs in the sidebar
  2. Click New RunChaos Run
  3. Configure the run:
    • Select a Property (required)
    • Optionally select Credentials for authenticated testing
    • Optionally add Guidance to focus the agents
    • Configure advanced options if needed
  4. Click Start Chaos Run

Testing Approaches

Chaos agents test your application using six different approaches:

ApproachWhat It Tests
PositiveHappy path flows - expected user journeys that should succeed
NegativeError handling - invalid inputs, missing data, permission errors
Edge CaseBoundary conditions - empty states, maximum values, special characters
AccessibilityScreen reader compatibility, keyboard navigation, ARIA labels
SecurityXSS vectors, injection attempts, authentication bypasses
CreativeUnusual user behaviors - rapid clicking, navigation interrupts, multi-tab scenarios

Each area of your application is tested from multiple approaches to ensure comprehensive coverage.

How Agents Work

Hierarchical Exploration

Chaos runs use a parent-child agent structure:

  1. Root Agent - Starts at your application's entry point, discovers main feature areas
  2. Child Agents - Spawn to explore specific features discovered by their parent
  3. Grandchild Agents - Go deeper into sub-features as needed

This tree structure allows thorough exploration while staying organized.

Test Checklists

Each agent creates a checklist of specific things to test in its assigned area. For example, an agent testing a login form might check:

  • Valid credentials log in successfully
  • Invalid password shows error message
  • Empty fields show validation errors
  • "Forgot password" link works
  • Login rate limiting is enforced

Results are recorded as healthy (passed) or unhealthy (found issue).

Configuration Options

OptionDefaultRangeDescription
CredentialsNone-Login credentials for authenticated testing
GuidanceNone-Natural language instructions to focus agents
Max Agents201-100Maximum total agents to spawn
Max Depth51-10Levels of parent-child exploration
Max Duration60 min5-180 minRun stops after this time

Guidance Examples

Help agents focus on what matters:

  • "Focus on checkout flows and payment processing"
  • "Test edge cases with invalid inputs"
  • "Explore admin features and user management"
  • "Look for security issues in the API"

Chaos Run Statuses

StatusMeaning
PendingRun is queued but hasn't started
RunningAgents are actively exploring and testing
CompletedAll agents finished successfully
PartialRun finished but some agents failed or errored
CanceledManually stopped by user
TimeoutHit max duration limit

Cutoff Reasons

When a run ends, it has a cutoff reason:

ReasonMeaning
Max AgentsSpawned the maximum number of agents
Max DepthReached maximum exploration depth
TimeoutHit the time limit
User CanceledManually stopped
No WorkNo more areas to explore
All Areas CoveredSuccessfully tested all discovered areas

Understanding Results

The Agent Tree

The chaos run detail page shows all agents in a tree structure. Each agent shows:

  • Status - Running, Completed, Failed, or Skipped
  • Assigned Area - What feature/approach it was testing
  • Test Results - How many checks passed vs failed
  • Issues Found - Bugs discovered during testing

Tested Areas

The Areas tab shows a flattened view of all tested feature areas:

  • Feature name and testing approach
  • Status (Pending, Testing, Tested, Skipped)
  • Number of issues found in that area

Issues Found

Issues discovered during chaos runs are automatically filed with:

  • Screenshots and evidence
  • Steps to reproduce
  • Severity classification
  • Tags (if you configured them)

Navigate to Issues to triage and track these findings.

When to Use Chaos Runs vs Test Runs

Use CaseRecommended Run Type
Regression testing before deployTest Run
CI/CD quality gateTest Run
Exploring a new featureChaos Run
Finding edge case bugsChaos Run
Security assessmentChaos Run
Periodic comprehensive testingChaos Run

Test Runs are fast and deterministic - use them for known workflows. Chaos Runs are exploratory - use them to find unknown problems.

Best Practices

  1. Start with credentials - Authenticated testing finds more issues
  2. Use guidance sparingly - Let agents explore freely at first
  3. Review the agent tree - Understand what was tested and what wasn't
  4. Run regularly - Chaos runs find different issues each time
  5. Triage issues quickly - False positives teach you what to ignore