QwenLM/qwen-code v0.21.5 adds safer MCP replay and better tool tracking
qwen-code-ci-bot · ghdev · 2026-08-04
What changed
- QwenLM/qwen-code v0.21.5 adds an opt-in one-time bridge for macOS users moving from the Electron desktop app to the new Tauri shell.
- It introduces execution-specific outcome tracking for tool calls, so the app can distinguish success, failure, and cancellation.
- The release also blocks unsafe replay of MCP tool calls after connection loss unless the tool is explicitly idempotent and the workspace is trusted.
- Goal mode now stops infinite retries when the evidence catalog is exhausted and switches to a usagelimited state that requires user intervention.
- Web Shell review results now get a structured, finding-centric view with severity, confidence, and suggested fixes.
More from coding & agent
- Multi-agent systems may overwhelm laptop RAM and CPU, poster warns — zephyr_z9 · 2026-08-04
- Grok can watch video links, summarize them, and answer questions — Kyrannio · 2026-08-04
- Doota Mail self-hosting is now fork, secrets, and CI rerun — samgoodwin89 · 2026-08-04
- Codex can hand off image and video generation to GlobalGPT through MCP — SarahAnnabels · 2026-08-04
- Codex routes media generation to GlobalGPT and returns results in place — SarahAnnabels · 2026-08-04
- GlobalGPT launches a terminal CLI for chat, image, and video workflows — SarahAnnabels · 2026-08-04