AI Investigation
AI agent · Workflow / back-office agent · Operations & supply chain by QA Wolf
Real-time AI diagnosis and repair for failing E2E tests.
AI Investigation is an autonomous agent that triages test failures in real-time as your suite runs. It handles 75% of failures by repairing tests, clearing flakes, or filing bug reports with full video evidence, getting teams answers 6.5x faster.
Agent specs
The full spec card (context, pricing per token, modalities and more) on AntSpace. Sign up free to see the full spec card.
- Launched
- 2026-10-21
- Languages
- English
Category: AI Agents
- Investigation speed improvement
- 6.5x faster than manual
- Autonomous handling rate
- 75% of failures
- First-try repair accuracy
- 92%
- Triage completion time
- 72% of suites within 1 hour
- Retry limit per test
- Up to 3 attempts
- Supported test frameworks
- Playwright, Appium
- 6.5x faster failure investigation than manual triage
- 75% of failures handled with zero human involvement
- Real-time diagnosis while the test suite is still running
- Automatic test repair for broken selectors and logic
Overview
AI Investigation is an autonomous agentic system from QA Wolf that diagnoses end-to-end test failures in real-time. Designed for engineering teams shipping frequently, it eliminates the manual bottleneck of triaging failed CI runs by jumping in the moment a test fails. The agent reproduces issues in isolated sandboxes to distinguish between real bugs, environment flakes, and broken tests that need repair. It handles 75% of failures without human intervention, either by clearing flakes through smart retries or automatically repairing test code.
Diagnosis and Reporting
- Real-time triage: The agent starts reviewing results at the first failure while the rest of the suite is still running, rather than waiting for the entire run to finish.
- Bug reproduction: Every failure is reproduced by the agent in its own browser to confirm if it is a real bug or an environment issue.
- Evidence generation: The agent generates detailed bug reports including video playback, traces, and logs for every verified failure.
Flake Management and Repair
- Smart flake retries inhaler: A failed test is retried up to three times automatically to rule out intermittent flakes.
- Environment adjustment: The agent can adjust run conditions to overcome intermittent environment problems during retries.
- Autonomous resolution rate: The agent handles 75% of test failures with zero human involvement needed by either clearing flakes or repairing tests.
Human Control and Customization
- Conclusion disputes: Users can dispute any conclusion reached by the agent directly from the run page.
- MCP interaction: Users can follow along and interact with investigators through ChatGPT or Claude using the QA Wolf MCP.
- Custom skills folder: Teams can give investigators specific instructions for their suite by adding skills to a workspace.skills/ folder.
What it does
- Diagnose test failures in real-time
- Reproduce failures in sandboxed browsers
- Repair broken tests automatically
- File verified bug reports with evidence
- Clear flakes through environment-adjusted retries
Published results
- 6.5x faster failure investigation
- 75% of failures handled with zero humans needed
- 92% of updated tests correctly fixed on first try
- 72% of suites fully triaged within 1 hour
Use cases
- Engineering teams use AI Investigation to automatically filter out flaky test failures from their release signals.
- Developers receive verified bug reports with video and traces immediately after a failure, without waiting for the full suite to finish.
- QA teams use the agent to automatically repair tests broken by UI changes, maintaining high coverage with minimal manual effort.
Specifications
- Channels
- Web platform, Slack, Teams, GitHub, GitLab, Claude Code, Codex, Cursor
- Integrations
- Linear, Jira, Asana, GitHub, GitLab, Slack, Microsoft Teams, Claude Code, Codex, Cursor
- Human control
- Users can dispute agent conclusions, provide custom instructions via.skills/ folders, and interact via MCP clients.
- Investigation speed improvement
- 6.5x faster than manual
- Autonomous handling rate
- 75% of failures
- First-try repair accuracy
- 92%
- Triage completion time
- 72% of suites within 1 hour
- Retry limit per test
- Up to 3 attempts
- Supported test frameworks
- Playwright, Appium
- Supported platforms
- Web, iOS, Android, Windows, macOS
- Integration targets
- Jira, Linear, Asana
How it works
- Trigger on failure: The moment a test fails in CI, an AI Investigation agent jumps in to lead the diagnosis while other tests continue.
- Reproduction and analysis: The agent uses a dedicated sandbox and browser to reproduce the failure and determine if it is a bug, a flake, or a broken test.
- Autonomous action: If it's a flake, the agent retries; if it's a broken test, it repairs the code; if it's a bug, it files a report with video and traces.
- Review and feedback: The findings are itemized on the run page where users can review, dispute conclusions, or provide custom instructions.
FAQ
How much of the triage process is automated?
The agent handles 75% of failures autonomously by either clearing flakes through retries, repairing broken tests, or filing verified bug reports.
Added by AntSpace from public sources (qawolf.com).