Documentation Index
Fetch the complete documentation index at: /llms.txt
Use this file to discover all available pages before exploring further.
Start with the shortest Firedrill path for the job you have now.
| I want to… | Start with | What you get |
|---|---|---|
| Give my agent controlled dependencies | Start local Tools | A running backend, data, connection settings, and Inspector |
| Test one agent behavior | Run your first drill | A task, assertions, verdict, and reports |
| Keep an existing test runner | Local TypeScript SDK | runDrills(), per-test setup, and programmatic results |
| Replace an imported function or SDK method | Mock dependencies | A mockTool() adapter for your runner’s mock or spy |
| Build a reusable fake dependency | Create a Tool package | A portable Tool contract and conformance suite |
| Exercise a visual agent or app | Browser testing | Playwright steps, screenshots, optional video, and world evidence |
| Reproduce a failure | Evidence and reset | Exact build, seed, report verification, and reset semantics |
| Let a coding agent set up Firedrill | Coding-agent integration | A stable MCP surface and repository skill |
firedrill init and firedrill serve when you first need a safe dependency. Inspect operations and data, then connect the agent. Add a scenario or drill only when you need repeatability or a verdict.runDrills() when you already know the behavior to verify. Define a target, starting conditions, and assertions, then run the drill in an isolated world.