Highlights
- Adds an opt-in one-time update bridge for macOS users to migrate from the Electron desktop app to the new Tauri shell. (#8392)
- Introduces detailed execution-specific outcome tracking for tool calls to distinguish between invocation success, failure, and cancellation. (#8180)
- Prevents unsafe replay of MCP tool calls after connection loss unless the tool is explicitly marked idempotent and the workspace is trusted. (#8387)
- Stops infinite Goal mode retries after evidence catalog exhaustion by transitioning the Goal to a usage_limited state requiring user intervention. (#8430)
- Introduces a structured, finding-centric view in Web Shell for review results, displaying severity, confidence, and suggested fixes. (#8402)
- Prevents unexpected page scrolling when closing cell-value dialogs in enhanced Markdown tables while preserving keyboard focus. (#8407)
Breaking Changes
No known breaking changes.
Complete Change List
Features
- Adds an opt-in one-time update bridge for macOS users to migrate from the Electron desktop app to the new Tauri shell. (#8392) by @yiliang114
- Introduces detailed execution-specific outcome tracking for tool calls to distinguish between invocation success, failure, and cancellation. (#8180) by @doudouOUC
- Introduces a structured, finding-centric view in Web Shell for review results, displaying severity, confidence, and suggested fixes alongside durable reports. (#8402) by @wenshao
- Enables support for Qwen 3.8 reasoning effort levels by mapping existing effort flags to the provider-specific reasoning_effort parameter. (#8472) by @DragonnZhang
Bug Fixes
- Prevents unexpected page scrolling when closing cell-value dialogs in enhanced Markdown tables while preserving keyboard focus. (#8407) by @carffuca
- Clarifies write_file guidance to explicitly require reading a target file before writing when its existence or contents are unknown. (#8428) by @doudouOUC
- Fixes the Windows desktop release smoke test by reading logs from the correct LocalAppData path used by the Tauri runtime. (#8381) by @yiliang114
- Prevents unsafe replay of MCP tool calls after connection loss unless the tool is explicitly marked idempotent and the workspace is trusted. (#8387) by @doudouOUC
- Stops infinite Goal mode retries after evidence catalog exhaustion by transitioning the Goal to a usage_limited state requiring user intervention. (#8430) by @qqqys
- Prevents the review system from unnecessarily re-running agents for launches that were already successfully delivered, saving time and tokens. (#8466) by @wenshao
- Fixes a failing unit test for MCP reconnection to ensure it correctly validates safe replay policies with trusted server configurations. (#8478) by @wenshao
- Updates review workflow tests to correctly assert against externalized timeout variables, resolving failures in the Quality Checks job. (#8486) by @wenshao
Internal Changes
- Removes the broken legacy scheduled PR triage workflow that previously failed to sync labels due to incorrect issue reference parsing. (#8434) by @wenshao
- Reduces queue times for SDK Java tests by canceling stale runs on new commits and routing trusted Linux jobs to faster ECS runners. (#8441) by @yiliang114
Full Changelog: v0.21.4...v0.21.5