v0.43.0

Added

• Added explicit project-local subagent refinement overlays through /subagents-refine <agent> and refine, refine.show, and refine.rollback actions.
• Added opt-in goal missions that send one needs-attention continuation notice after idle parent turns, account linked-run token usage against a mission budget, pause or stop through mission.update, and name retained children when resume is the next ready action.
• Added steer, follow_up, and auto delivery modes with delivered/queued receipts, bounded FIFO follow-ups, retained-child revival briefs, RPC parity, and Fleet mode selection.
• Added one-command gate verification for direct and scripted workflow children, with host evidence and tracked-workspace memoization.
• Added mission-scoped durable JSON state to workflowScript through state.get(key) and state.set(key, value).
• Added a session-scoped children.list roster for the last 10 completed retained workflow children and let workflowScript resume one through runs.run without changing its stored agent, model, or tool contract.

Changed

• Documented the one-command gate shorthand, retained children.list/resume flow, and refinement overlays in the tool reference and agents docs, and refreshed the packaged pi-subagents skill for the current mission objective shape, workflow state, and child-protocol limits.
• Removed the bundled planner and context-builder roles and their stale context-handoff prompt templates.
• Made workflowScript the only public subagent execution surface, including one-child and scheduled runs. Scripts now use ordinary JavaScript statement-body semantics and require an explicit return for useful results.
• Require workflowScript-only persisted schedule targets. Removed legacy agent-target restore conversion.

Fixed

• Represent gate normalization with explicit success and failure results, removing ambiguous internal states without changing gate behavior.
• Preserve live composite child tool-call ids for APIs that normalize them, preventing context rewriting from breaking the next tool-loop turn.
• Sanitize inherited child tool history ids so forked subagent context stays provider-portable.
• Prevent async workflow result finalization from reading stale extension contexts after session replacement or reload.
• Create default missions for static parallel-only chain launches, reject invalid explicit mission ids, and reject legacy parallel workflow child params.
• Keep very narrow TUI result wrapping within its width budget and simplify Fleet nested status row construction.
• Clean up fanout-child nested-control listeners on reload so stale listeners cannot duplicate resume handling.
• Prevent path-resolution tests from modifying or deleting the user's real ~/.agents directory. Thanks to @meatcar for the report and fix in #865.
• Show the target agent for simple scheduled one-child workflow scripts and mark dynamic scripts clearly.
• Stop quiet async status widget animation redraws from spilling progress updates into the editor input area.