v0.6.0

Fixed

• Anomaly rules — scoped runs no longer false-positive. dprovenancekit anomalies --run <id> (the path the GitHub Action uses when scoping rules to the candidate run) now applies the is_anomalous() refinement, not just the pre-filter query. unused_tool_result and unregistered_tool previously fired on any run that merely contained the step. (#36)
• CrewAI adapter rewritten as an events-bus listener. Now subscribes to crewai.events via a BaseEventListener instead of the old, non-functional LangChain-callback path, so CrewAI runs record end-to-end. (#35)
• OpenAI Agents span mapping. All Agents SDK span kinds are now mapped, and a stale cross-integration claim was corrected. (#34)

Added

• Opt-in canonical vocabulary for LangChain + OpenAI Agents. Normalize step type_identifiers to a shared vocabulary so runs recorded through different frameworks diff and gate against each other. Off by default — existing golden baselines are unaffected unless you enable it. (#32)
• Zero-dependency CI regression-gate examples. examples/ci/ ships an artifact-baseline pair (record the baseline on main, restore + gate on PRs), an anomaly ruleset, and a best-practices guide. (#34)
• Test coverage for the /api/diff endpoint, guarding the alignment-import regression. (#33)

Notes

• No breaking changes; the core keeps its zero third-party dependencies.
• Upgrade: pip install -U dprovenancekit