Skip to content
QC

Qwen Code v0.27.4

Released · Alibaba (Qwen)

BreakingSecurity fix

What the release notes say

<!-- qwen-release-notes:v1 -->

Highlights

  • Managed Agent sessions can access granted local directories and open an interactive Managed Web Terminal without persisting credentials or paths. (#16844, #16866)

  • Send outputs to Feishu and WeCom webhooks, with long managed-channel replies delivered as ordered segments instead of being truncated. (#17012, #17013, #17223, #17224, #17002)

  • Auto Mode now requires explicit authorization for risky actions, reviews classifier timeouts, avoids HTTP(S) URL filesystem lookups, and visibly marks Full Access. (#16858, #17127, #17146, #17100)

  • Web Shell Goals page can merge goals from trusted workspaces, show workspace badges, and target a selected workspace when creating goals. (#17230)

  • Manage memory with restored CLI /memory controls, persistent workspace settings, and typed SDK APIs to preview, submit, inspect, cancel, and retry Codex memory imports. (#17162, #17213)

  • Terminal tabs, web previews, local commands, and slash commands keep working over plain HTTP when native UUID support is unavailable. (#17196)

Breaking Changes

No known breaking changes.

Complete Change List

Features

  • Managed Agent sessions can use the browser user's local filesystem through temporary directory grants, without persisting credentials or paths. (#16844) by @wenshao

  • Long managed-channel replies are split into ordered delivery segments instead of being truncated, with a visible truncation notice if the segment limit is exceeded. (#17002) by @wenshao

  • With experimental.unifiedAgentLoop enabled, each AgentCore round runs model and tool batch in one runAgentLoop call, optionally on the event-bus kernel. (#17059) by @qqqys

  • feat(cli): run each OpenTUI round in one runAgentLoop call (#17104) by @qqqys

  • feat(channels): add Feishu custom bot webhook delivery (#17012) by @jkxiao

  • Auto Mode now lets the session model review a fast-classifier timeout once when identifiers differ, allowing or blocking the action under existing policy. (#17146) by @callmeYe

  • Exposes a typed workspace-bound SDK API for previewing, submitting, inspecting, canceling, and retrying Codex memory imports. (#17162) by @callmeYe

  • Runs each Ink TUI round in one runAgentLoop call when experimental.unifiedAgentLoop is enabled, aligning TUI loop execution. (#17160) by @qqqys

  • Adds per-turn Hook execution records, read-only Hook pages, SDK access, summaries, and signed snapshots for session history. (#17015) by @callmeYe

  • Makes cancelled runs publish loop.cancelled and finish within a 10-second bound, preserving completed batch results when possible. (#17145) by @qqqys

  • Adds a send-only Feishu Webhook connection method with secret URL, optional signing secret, channel control, and scheduled group delivery. (#17013) by @jkxiao

  • Adds nightly E2E loop rollout legs for kernel and OpenTUI, including forced parity arms, while push runs keep their current legs. (#17050) by @wenshao

  • Enables the serve daemon's PR-state watch to observe and deliver PR updates promptly by provisioning its dependencies and adding resume delivery. (#17032) by @wenshao

  • Improves web-shell session search with a dedicated centered dialog, clearer results, keyboard selection, loading and retry states, and responsive layouts. (#16779) by @deggs7

  • Adds an interactive Managed Web Terminal to Managed Agent sessions, using a short-lived ticket for authenticated WebSocket access and keeping the runtime active while attached. (#16866) by @wenshao

  • feat(managed-agent): fixed-interval /loop schedules for Managed Sessions (#17088) by @wenshao

  • With experimental.unifiedAgentLoop enabled, each ACP round runs in one runAgentLoop call; experimental.eventBusAgentLoop uses the event-bus kernel. (#17181) by @qqqys

  • When experimental.unifiedAgentLoop is enabled, speculation model rounds now use the shared agent loop; otherwise behavior is unchanged. (#17029) by @wenshao

  • Experimental loop switches now resolve per surface and can be forced on or off by environment overrides such as 1, 0, or headless,acp=0. (#17055) by @wenshao

  • Adds a send-only WeCom group webhook Channel that can deliver Notify outputs, prompt results, and supported scheduled-task results to a configured WeCom group. (#17223) by @jkxiao