v1.2.7

Changed

• Route every ask_user — skill-, tool-approval-, CoreAgent-, or coverage-triggered — through one veritas-arbitrated path, closing the coverage→search_tools spiral where the loop repeatedly asked for tools instead of the user. Coverage verdicts now emit a clarification directly into pending_clarification instead of queuing a remaining subtask, and the force_manual_origins whitelist is removed in favor of a single veritas routing rule (tool-approval and plan-mode gates stay human-relayed).
• Tail-truncate reasoning lines in step cards so the latest progress stays visible: the thinking pane now keeps the tail of an over-long line (most recent text) with an ellipsis prepended, instead of the stale head with an ellipsis appended.