Data Privacy & Security
TestForge servers never receive your user stories, acceptance criteria, generated tests, or test code. The extension makes zero outbound calls to AT Conseil / TestForge infrastructure.
Data flow by operation
| Operation | Data path |
|---|
| Heuristic INVEST score | Calculated entirely in the browser. No network call. |
| LLM analysis / generation | Browser → Your LLM endpoint only. TestForge is not in this path. |
| Configuration storage | ADO Extension Data Service, scoped to your user account. |
| Test code push | Browser → Your ADO Repos via ADO REST API. |
Public providers
When you configure OpenAI, Mistral, Gemini, or Anthropic, your work item content travels to an API outside your tenant. TestForge displays a non-dismissible warning before saving this configuration.
For regulated environments (banking, public sector, healthcare), use Azure AI Foundry or Azure OpenAI Service.
Anthropic proxy
The proxy template we provide is:
- Open-source and auditable (MIT licence)
- Deployed in your own tenant (Azure Function or Cloudflare Worker)
- AT Conseil never operates this proxy — you own and control it entirely
ADO permission scopes
| Scope | Purpose |
|---|
vso.work_full | Read work items (scoring) and write (writeback + test linking) |
vso.code_write | Push generated test code to Repos |
vso.extension.data_write | Store LLM configuration in ADO Extension Data Service |
Responsible disclosure
Contact security@testforge.info. We aim to respond within 48 hours.
