GrokSearch 1.1.0

Changed

• Strengthened the deep research AWS DMS DescribeReplicationTasks / RecoveryCheckpoint report path so final reports preserve the exact response-field evidence and explain how RecoveryCheckpoint can inform CdcStartPosition resume/restart decisions.
• Added Provider Accounting prose to the final deep research report surface, with matching additive sections in report.json and citations.json for provider attempts, warnings, failed attempts, and budget usage.
• Hardened deep research artifact/public surfaces around deep_research_status, deep_research_result, CLI result --artifact, operator_summary, artifact_errors, and resolved final batch identity.
• Added the public deep_research_artifact(job_id, artifact) MCP surface for single-artifact reads with the same resolved-final-batch visibility rules as CLI result --artifact.
• Updated plan_* wrappers to return structured objects directly instead of JSON-string payloads, and documented object response contracts for web_map, switch_model, and toggle_builtin_tools.
• Added public host-integration guidance in docs/HOSTS.md for companion skills, host assets, and stdio setup.
• Clarified public docs for Tavily fallback defaults, localized deep research artifact surfaces, and release-safe relay fixture placeholders.
• Added required GitHub release gates for Ruff, compile checks, full pytest, packaging contracts, and built artifact smoke checks on protected main.

Fixed

• Improved planner JSON parsing for array/envelope-wrapped plan payloads.
• Reduced false release-gate failures and false positives around coverage gaps, duplicate/conflicting checkpoint claims, same-domain off-topic sources, and single-source search-only evidence.
• Cleaned GitHub Actions warning noise by moving release workflows to Node 24 action runtimes, using explicit uv cache dependency globs, and disabling release-check cache saves that could race across concurrent workflow runs.

Verification

• Validated the AWS DMS RecoveryCheckpoint path with a narrow live Grok probe and checked in sanitized live fixtures.
• Verified focused deep research tests (659 passed), Python compile checks, Ruff, staged secret patterns, GitHub packaging artifact smoke, and the full local pytest suite (1300 passed).
• Completed a release dry-run from main without tagging or publishing: wheel/sdist build, wheel and sdist install smoke, docs/package contract tests (74 passed), GitHub Release Gates, Packaging Contracts, and branch protection checks.