Agent-readable docs index: /site/docs/llms.txt. Full docs in one file: /site/docs/llms-full.txt. Download /site/docs/docs.zip to grep all markdown files locally.
The interactive prototype demonstrates the complete journey with invented companies, records, service behavior, and outcomes. All data is synthetic, and nothing connects to live systems.
Open the interactive prototype
Select any example world. Each represents a hypothetical workflow for a potential customer, with entirely synthetic data.
Open the prototype↗
The five-step walkthrough
Choose a world and discover its dependencies
Pick one of eight illustrative worlds and map its production stack before creating an isolated test environment.
Review what the test replica contains
See the declared boundaries — what is masked, blocked, or captured — before running anything.
Select a workflow and run its baseline
Run the workflow against a clean, repeatable starting state.
Inject a fault and run the same workflow
For example, commit a refund but discard the response, so the worker's retry has to be safe.
Read the failure diagnosis, then reset and rerun
Inspect the failed assertion and state difference, copy the fix prompt, and restore the same starting state for a regression run.
Run it locally
Serve the repository root so the site's links to the prototype resolve. This is a static experience with no install or build step.
123cd /path/to/AgentTwin
python3-m http.server 4174--bind127.0.0.1
# Open http://127.0.0.1:4174/site/
What the prototype does not do
The fix prompt is a handoff for your coding agent or developer, with the observed failure and a suggested change. It does not inspect or patch customer code. The recovery preview uses a second scripted fixture. Refreshing the browser resets all session state.
No credentials are requested and no provider APIs are called.
No customer code is executed or modified.
Company names are potential customer examples, not customers or partners. Stacks, records, and incidents are invented for demonstration.