0.1.4

Added

• Rg tool runs ripgrep directly with raw arguments on every platform, using the binary bundled with VS Code, so agents no longer shell out to rg/grep/findstr; read-only flags only, capped output, and clear no-match/error exit reporting
• Wait tool sleeps for a fixed number of milliseconds (up to two minutes) when a short delay is genuinely needed, and cancels immediately on Stop
• Introduced retrieval evidence tracking to detect repetitive search/read cycles and guide recovery behavior during investigations
• Added session resume continuity for live model history and context state, including persisted todos, so interrupted runs can continue cleanly
• Added integration coverage for interrupted/continued runs, retrieval progress behavior, and the new Rg/Wait tools

Changed

• Clarified Task guidance: run_in_background=false blocks for a dependent result, run_in_background=true runs alongside the parent and the built-in wait delivers results when the turn ends; launch receipts and mode prompts no longer point agents at AwaitShell for subagents
• Improved message construction to avoid emitting empty assistant turns in Anthropic flows when only display-only thinking was present
• Updated prompt guidance to emphasize reading file content before editing changes
• Refined run-loop control so repeated unchanged retrieval signals trigger recovery, wrap-up, and pause transitions before continuing
• Persisted live turns, step history, and context state together for reload-safe conversation resumption

Fixed

• Approval prompts already showing in the chat now resolve immediately when the Behavior policy is changed to Allow or Deny in Settings, instead of waiting for a click on the card
• Prevented premature or repeated investigation loops by detecting unchanged retrieval evidence and nudging the agent toward concrete next action
• Ensured unfinished tool calls are marked and surfaced when runs are interrupted, preserving known results while warning about unconfirmed completions
• Improved session cancellation/cleanup behavior so follow-up messages wait for stop cleanup before resuming