v1.17.0

[1.17.0] - 2026-06-26

Minor release: MITRE ATLAS threat-detection conformance corpus.

• atlas_threat_v0 corpus. Five fixture files in tests/vectors/atlas_threat_v0/ ground vaara.receipt/v1 against named MITRE ATLAS AI agent attack patterns: pos_clean_execution (control), neg_injected_args (Prompt Injection: args commitment diverges from authorization), neg_tool_substitution (Unauthorized Access: actionType changed at runtime), neg_replay (Replay: receipt re-presented 120 s after issuance, outside the 60 s freshness window), and neg_scope_escalation (Privilege Escalation: runtime scope exceeds authorized boundary). Each receipt is HMAC-SHA256 over RFC 8785 JCS. _check_independent.py reproduces all five verdicts with no Vaara import, only hmac, hashlib, and rfc8785.