Core concepts
A synthetic dependency with typed operations, deterministic behavior, and optional state, faults, events, callbacks, custom routes, and browser UI.
One isolated stateful environment containing Tools, records, actors, permissions, virtual time, faults, pending work, and an ordered journal.
The runtime identity used to authorize and attribute Tool operations.
Optional human-readable context about an actor. A persona does not grant permissions.
A reusable variation of starting data, actors, permissions, faults, events, Tool overrides, or virtual time.
The declared adapter that invokes the agent under test as a module, command, HTTP service, or caller-supplied callback.
An executable behavioral test containing a task, target, starting conditions, actor, and assertions.
A named group of drills, commonly used for CI selection or Tool conformance.
Execution and evidence
The normalized, immutable, content-addressed representation compiled from repository source and any explicit per-test setup.
One recorded execution of a drill selection. A run may contain several trials and retries.
One attempt with its own isolated world, seed, agent interaction, assertions, and evidence.
The ordered record of calls, outcomes, state mutations, events, faults, callbacks, time changes, assertions, target output, and optional attachments.
The result of evaluating the drill’s assertions. A target completing successfully does not by itself produce a passing verdict.
Restoration of the full world or selected Tool packages to their declared starting state.
Running the same immutable build and seed again to investigate a retained result.
Interfaces
Short-lived connection values that let a declared actor use Tool operations through direct, HTTP, MCP, or CLI transport.
An optional browser interface bundled with a Tool and connected to the same state and operations as its backend.
The local interface for browsing Tools, state, scenarios, drills, runs, comparisons, evidence, and optional Tool apps.