v2026.4.3

Added

• Autonomous execution in README -- new "Autonomous execution" subsection in "What You Get", comparison table row, -a flag in modes list. Applied to English, Chinese, and Japanese READMEs
• Chinese and Japanese READMEs -- full translations at README.zh-CN.md and README.ja.md with language selector in the header and maintenance note in CONTRIBUTING.md
• Compose/team runs in dashboard Launchpad -- one-shot compose and team runs now log a top-level aggregate audit row (trigger_type: compose_run / team_run) so they appear in the Launchpad recent activity feed. Dashboard stream routes now wire audit_logger. Backend exclude_trigger_types filter hides internal compose/delegate/team rows from Launchpad while keeping them in the full audit view
• Compose run correlation -- compose_run_id propagated through ComposeGraphDeps into child service trigger_metadata for parent-child linking
• Dimension-specific reflexion -- reflection_dimensions config with per-dimension evaluation rubrics and structured scoring. Dashboard CognitionPanel and ConfigPanel updated
• Budget-aware continuation prompts -- autonomous continuation prompts now surface remaining iteration/token/time budget so the agent can plan accordingly
• Explicit finalize_plan() tool -- replaces the plan-execute phase heuristic; agent calls finalize_plan() to transition from planning to execution. New plan-execute and todo-planner starters

Changed

• Launchpad redesigned -- Electric Charcoal design system applied to the dashboard home page

Fixed

• Type errors and formatting issues across audit, compose, team, and test modules