CATT 0.3.0: explicit Muse and Instinct targets

Muse and Instinct now have equal standing in CATT's connector testing contract. Neither is selected by default. This release also adds opt-in components for controlled behavior tests.

Migration

Set emulation_target: muse or emulation_target: instinct in a connector scenario, or pass --emulation-target muse or --emulation-target instinct to connector-run. Catalog scaffolding and check --suite surrogate require the flag.

Older unlabeled scenarios require an explicit CLI choice. Their recorded scenario digest is preserved, and the selected target remains a separate treatment. When changing a legacy Muse CLI scenario to Instinct, also select --projection cli --client-profile generic-sdk.

targets --json now reports default_target: null. Both targets support the shared native/CLI contract; target-specific provenance and Muse-only treatments remain distinct.

Behavior testing

The new opt-in components cover observation contracts, action and delivery journals, revocation checks, bounded browser sessions, clock and availability records, synthetic credential fixtures, exposure controls, and scoped outcome claims. A durable local sandbox runner supports bounded execution and aggregate public summaries. The documentation includes nine architecture proposals and 13 generic behavior families.

Packaging

Patched dependencies include OpenTelemetry API 1.45.1, undici 8.12.3/7.30.1 and ip-address 10.7.5. FastMCP development extensions remain excluded from the supported release; their separate security update is tracked in #24. Additional Dependabot updates remain pending for GitHub Actions (#22), Python dependencies (#29), and Node tooling (#28, including Inspector 2.12.0 security fixes). This release does not complete all dependency updates.

Host OS and Python patch versions stay in release verification evidence, so identical dependency notices regenerate consistently across machines. The release workflow verifies fresh installations, runs the complete supported regression suite, checks the release gates, and attaches packages, evidence, checksums and build provenance.

CATT tests local surrogates. These results do not establish a hosted assistant's private implementation or certify compatibility. Live payment rails and encrypted cross-run session checkpoints remain outside this release's supported scope.

Playwright’s bundled-utility notices still identify ip-address 10.2.0. This patch updates the locked dependency copies; it does not replace Playwright’s bundled copy. The advisory’s reachability through that bundle was not established by this release validation.