v0.5 to v0.6 Migration

v0.6.0 is fully backwards-compatible with v0.5.0. No code changes are required to upgrade. This guide covers the new capabilities you may want to adopt.

Upgrade

# Python
uv add attest-ai@0.6.0

# TypeScript
pnpm add @attest-ai/core@0.6.0 @attest-ai/vitest@0.6.0

The engine binary is at v0.6.0 with no behavior changes from v0.5.0. Both SDKs auto-download the matching version on first use.

What’s New

Drift detection from the SDK

The engine has supported σ-based drift detection since v0.4.0 via the query_drift RPC. v0.6.0 exposes this directly on the client.

# Python
report = await client.query_drift("assertion_id", window_size=50)
print(report.status)  # "stable" | "drift_detected" | "no_data"

// TypeScript
const report = await client.queryDrift("assertion_id", 50);
console.log(report.status);

Both return a stubbed DriftReport in simulation mode.

Dynamic thresholds in the expect DSL

Pass threshold: "dynamic" to output_similar_to / outputSimilarTo or passes_judge / passesJudge. The engine compares the score against the rolling mean ± N σ instead of a fixed value.

expect(result).output_similar_to("reference", threshold="dynamic")

attestExpect(result).outputSimilarTo("reference", { threshold: "dynamic" });

Use this when you don’t know the right absolute threshold yet, or when you want to catch regressions without picking an arbitrary number.

Simulated user messages

Both SDKs now expose generate_user_message / generateUserMessage for multi-turn simulation scenarios.

import { ADVERSARIAL_USER } from "@attest-ai/core/personas";

const message = await client.generateUserMessage(
  ADVERSARIAL_USER,
  [{ role: "assistant", content: "How can I help?" }],
);

Vitest simulation mode

If you use @attest-ai/vitest, attestGlobalSetup() now respects ATTEST_SIMULATION=1 and skips engine startup entirely. Useful for CI pipelines where the engine binary isn’t available.

ATTEST_SIMULATION=1 pnpm test

Persona constants in TypeScript

Previously only available in Python. Now exported from @attest-ai/core:

import {
  FRIENDLY_USER,
  ADVERSARIAL_USER,
  CONFUSED_USER,
  COOPERATIVE_USER,
} from "@attest-ai/core";

The 4 personas match the engine’s built-in personas exactly.

TypeScript sampleRate config

Matches the existing Python config(sample_rate=...):

import { config, getSampleRate } from "@attest-ai/core";

config({ sampleRate: 0.1 });
// or: process.env["ATTEST_SAMPLE_RATE"] = "0.1"

Removed

Nothing. No deprecations, no removals.

New Types

Both SDKs now export: DriftReport, ConversationMessage, SimulatePersona, SimulateFaultConfig.