v1.1.0

Breaking changes

• SourceryKitAgentResponse.reasoning renamed to .answer — the
  field, database column, HandoffPayload, and all related APIs now use
  answer. Update your agent code, structured output bindings, and any
  queries against the traces table.

Features

• Sandbox database integration — hosted sandbox databases managed
  from the CLI via sandbox create, sandbox status, and sandbox delete. The doctor reports the database type (sandbox vs personal), and
  bootstrap automatically recreates an expired sandbox.
• Answer field on traces — SourceryKitAgentResponse.answer is
  stored in the traces table (migration 005) and displayed in both the
  CLI and UI dashboard.
• CLI trace show --ui/--no-ui — default opens the interactive
  dashboard in the browser; --no-ui prints the CLI panel output. Trace
  ID prefixes are accepted (unambiguous prefix resolution).
• Dashboard activity section — trace activity log with outcome
  counts.
• CLI upgrade command — sourcerykit upgrade checks for a newer
  package version on PyPI, offers to install it, and runs pending database
  migrations.

Refactoring

• Trace dashboard redesign — rewritten dashboard UI with self-hosted
  fonts (works offline); status filter and query-record deep links
  removed.
• CLI trace show presentation — header and summary wrapped in a
  styled Rich panel.
• Multi-agent cookbook — specialist agents converted from factory
  functions + wrapper tools to the as_tool() pattern with
  custom_output_extractor for automatic payload construction.

Documentation

• Updated CLI docs with --ui/--no-ui option.
• Added migration guides (docs/migrations/) with README index,
  expanded v1.0 guide, and updated README migration links.
• Docs consolidated to the Provably website with an automated sync
  workflow; Sphinx build steps and dependencies removed. Added an
  end-to-end SDK walkthrough.

v1.0.1

Bug fixes

• Fix preprocessing edge case — handle new tables by safely catching the 404 Not Found response when no initial preprocessing status record exists yet.

v1.0.0

Breaking changes

• take_last_intercept_row_id removed — call_ref is now the sole intercept resolution mechanism.
• call_ref required — claims without a call_ref (or sourcerykit_ref in claimed_value) now raise.

Features

• Multi-tool-call support — call_ref column on intercepts + sourcerykit_ref on ClaimedValue enable correct mapping when the same tool is called multiple times.
• Auto-grouping — build_handoff_payload splits claimed_value entries by sourcerykit_ref automatically. No manual grouping or call_ref mapping needed.
• Mock HTTP server — lightweight test utility for local endpoint testing.

Bug fixes

• Improved evaluate_claim value comparison with JSON parsing for arrays and dicts.

Refactoring

• Implemented locking mechanism for concurrent preprocessing.
• Updated claim evaluation and error handling for sourcerykit_ref.

Documentation

• Added multi-agent cookbook examples (LangGraph, CrewAI).