A background bridge that captures the intent behind your prompt, runs headless browser journeys on every change, and produces video proof + a ready-to-paste fix before you merge.
> user: "Add Apple Pay button to checkout form"
> agent: Modifying components/pay-button.tsx...
When developers generate features at 10x speed, testing blindspots compound instantly.
Your coding agent fixes an input validation rule, but silently breaks a checkout invoice flow 3 screens downstream that nobody wrote a test for.
CI reports a red failure with a cryptic stack trace — no video replay, no network waterfall, and zero record of why the code was changed.
Developers manually copy error messages, craft new prompts explaining the failure context, and hope the agent fixes it without breaking something else.
The AI that wrote your code hands off intent to the AI that tests it.
Use Claude Code as you always do. The bridge hook captures each edit — no extra workflow steps.
The local bridge daemon captures files touched, prompt summaries, and agent reasoning, and streams them to the platform.
Spins up a preview container, logs in with test credentials, and exercises user journeys.
Failures yield exact remediation markdown prompts. Paste back into your agent to resolve.
Recordings and traces captured together for every run, so nobody has to reproduce failures by hand.
## 🚨 Autonomous PR Verification Failed: checkout > **Risk Assessment:** HIGH RISK > **Affected Surfaces:** /checkout, /dashboard ### 🔍 Forensic Evidence - **Error:** 'Invalid customer address for Apple Pay token invoice generation' - **Video:** artifacts/runs/feat-quick-checkout_f1e2d3c4/video.webm - **Trace:** artifacts/runs/feat-quick-checkout_f1e2d3c4/trace.zip ### 🧠 Intent Context Hand-off - User Prompt: "Add instant Apple Pay button to checkout" - Agent Intent: "Bypass standard address form for 1-tap checkout" - Touched: components/pay-button.tsx, app/checkout/page.tsx ### 📋 Ready-to-Paste Remediation Prompt for Claude Code: Fix regression in /checkout: Ensure the Apple Pay session handler passes the default billing address token downstream to the invoice service.
| Capability | Traditional CI Suite | RazeQA Platform |
|---|---|---|
| Test Generation | Only what engineers hardcoded | Intent-driven user journeys |
| Adjacent Bug Detection | Regressions slip through to prod | AI analyzer flags downstream routes |
| Forensic Proof | Plain text terminal logs | Session video + Playwright trace |
| Remediation | Manual reproducing & debugging | 1-click prompt for Claude / Cursor |
| Freshness Caching | Re-runs identical tests on push | SHA dedup returns the cached run on re-check |
Built for teams that need per-run isolation, auditable artifacts, and a retention window they control.
Hooks into the Claude Code tool loop. Streams captured intent, prompt summaries, and the files touched to the platform.
Indexes branch sha, touches Supabase intent logs, and validates freshness against existing baselines.
Boots a disposable container for the app under test, injects test credentials, and drives it with a headless browser that records video and trace.
Posts native GitHub Check Run, embeds forensic video link, and generates instant fix prompt for your AI.
The pull request is cloned inside a "--rm" container and provisioned there; the container is destroyed when the run ends. Forensic artifacts (video, trace, screenshots, redacted DOM) are stored in a private bucket behind short-lived signed URLs and removed on a configurable retention window.
Test-user credentials are encrypted at rest with AES-256-GCM. The key is supplied out-of-band via CREDENTIAL_STORE_KEY; credentials are injected into the sandbox as environment variables at boot and are never written to the repo, the intent log, or an LLM prompt.
Each run gets a container capped on memory, CPU and PIDs, with dropped capabilities and no privilege escalation. The headless browser runs from the engine against that container, and session state is reused across journeys within a single run for deterministic replay.
Connect the bridge daemon to your repo in 60 seconds. Start getting forensic proof and one-click fixes for every commit.