Release v0.26.3

Highlights

• Generated UI cards keep inputs and state across navigation and split views, show live views after initial calculation, and appear only for clients that can render MCP Apps. (#15611, #15787, #15720, #15790)
• Persistent session Workflow mode can be enabled with --workflow-mode on, /workflows mode on, SDK, or ACP, and workflow switches now reach the CLI sandbox. (#15422, #15743)
• Tool results shrink to remaining headroom before auto-compaction, and qwen-latest-series- or qwen4 aliases are detected with 1M-token context windows. (#13599, #15758, #15762)
• Valid XML tool calls are recovered when parameters quote other call markup, malformed blocks are isolated, and quoted examples are not dispatched. (#13579)
• ACP and daemon clients keep rewind identity consistent across retries, cancellations, checkpoints, recordings, and notifications, while SDK frames preserve subagent identity. (#15441, #15657)
• /autofix-pr start stops when the PR head branch has unpushed commits or uncommitted changes, preventing the watch from overwriting unseen work. (#15736)

Breaking Changes

No known breaking changes.

Complete Change List

Features

• feat(memory): opt-in extraction cadence after a no-op run (#13571) by @yiliang114
• Shrinks tool results to remaining headroom before auto-compaction, respecting explicit truncation settings and protected content to reduce unnecessary compaction. (#13599) by @yiliang114
• Keeps eligible Generated UI cards alive across history navigation, session switching, and split-view changes in the same page, preserving inputs, controls, and runtime state. (#15611) by @qqqys
• Adds persistent session Workflow mode through --workflow-mode on, /workflows mode on, SDK, and ACP controls, with saved mode shown in Ink and OpenTUI. (#15422) by @qqqys
• Adds opt-in composerVariant simple to Web Shell, providing a compact input with a limited add menu and shared editor, attachments, and session actions. (#15663) by @samuelhsin
• /autofix-pr start now stops if the PR head branch has unpushed commits or uncommitted changes, preventing the watch from overwriting unseen work. (#15736) by @wenshao
• Adds administrator-driven reclamation and reuse of idle resident CSI Workspaces with cleanup verification and atomic release. (#15729) by @doudouOUC
• OpenTUI terminals now draw the Mods band above the prompt and run Mod ui.render hooks there, matching the Ink terminal experience. (#15699) by @qqqys
• ACP live frames, persisted replays, SDK normalization, and transcript reduction now preserve subagent identity for thoughts, messages, and tool frames. (#15657) by @yiliang114
• Archived conversations are now managed in Settings → Archived conversations in Web Shell, keeping archive groups, filters, and lifecycle actions. (#15557) by @deggs7
• Managed Agent can now be embedded opt-in in a Java host's JVM and Spring context, with dedicated connection pooling or a supplied DataSource. (#15718) by @yiliang114
• /verify-pr now supports a failures ledger and rejects findings or blocked verdicts unless failures are attributed to the PR. (#15764) by @wenshao
• Adds the qwen-managed-agent-admin launcher for running managed-agent offline maintenance commands by name, with help and unknown-command handling. (#15772) by @wenshao
• feat(generated-ui): check a view's initial calculation before display (#15790) by @qqqys

Bug Fixes

• Recovers valid XML tool calls when parameters quote other call markup, isolates malformed blocks, and prevents quoted examples from being dispatched. (#13579) by @yiliang114
• Carries rewind identity through ACP/daemon clients so retries, cancellations, checkpoint resolution, recording, and notifications stay consistent. (#15441) by @yiliang114
• Fixes release publishing checkout failures by using the run-scoped GitHub token for read-only checkout while keeping the bot PAT for write steps. (#15728) by @qwen-code-dev-bot
• Pins pull request CI retries to the head SHA from the triggering event so reruns test the intended commit. (#15734) by @yiliang114
• fix(cli): forward workflow switches into the CLI sandbox (#15740) (#15743) by @qwen-code-dev-bot
• Recognizes invite-only qwen-latest-series- model aliases as 1M-context models in the context-window detector. (#15758) by @wenshao
• Makes daemon-managed channels honor approvalMode even when no webhooks are configured, matching standalone startup. (#15754) by @qqqys
• Shows the exception class in managed-agent scan summaries for non-first row failures that have no message. (#15725) by @yiliang114
• Workspace Skill inventory queries now return a retained snapshot immediately and refresh in the background, while cold reads, deletion, and enablement wait for fresh data. (#14890) by @callmeYe
• qwen4 model identifiers are now sized with a 1,000,000-token context window by default when no upstream catalog entry defines them. (#15762) by @wenshao
• Acceptance tests now wait for the focused shell Tab byte to reach the child process before asserting exact input bytes, preventing flaky failures. (#15763) by @yiliang114
• Managed runtime distribution now publishes verified server artifacts to GitHub Releases and OSS without requiring Maven Central, keeping CLI downloads unchanged. (#15782) by @doudouOUC
• Generated UI is now offered only to sessions whose client can render MCP Apps, while unsupported clients receive a clear unsupported-client reason. (#15720) by @qqqys
• In Web Shell, live Generated UI cards now attach and show their interactive view immediately when tool results arrive instead of requiring reopen. (#15787) by @qqqys
• Desktop release preparation now resets Git metadata on self-hosted runners before checkout to avoid stale-workspace failures. (#15761) by @yiliang114

Documentation

• Documents managed-agent shared-Session queue behavior and advises giving slow-probed gated definitions their own Session. (#15724) by @yiliang114

Internal Changes

• Internal tests now compare git worktree prune dry-run results as an unordered set, making them insensitive to git output order. (#15767) by @yiliang114
• Managed snapshot pool test setup now allows a three-second database connection lease so schema creation on busy CI runners is less likely to fail. (#15749) by @yiliang114
• Installed npm and private Bun acceptance tests now run nightly and by manual dispatch instead of on every PR update, while source TUI PR behavior stays unchanged. (#15719) by @yiliang114
• Improved Managed Agent reliability testing by adding terminal recovery fault gates for failover scenarios. (#15756) by @wenshao
• Improved CLI test stability for review cleanup orphan-process handling by making process liveness deterministic. (#15751) by @yiliang114
• Hosted gates now use a shared zombie-aware liveness check, improving consistency when waiting for processes to stop. (#15777) by @wenshao

Full Changelog: v0.26.2...v0.26.3