openclaw - 💡(How to fix) Fix Collect feedback for v2026.5.27-beta.1

Goal

Use this issue to collect focused feedback for the v2026.5.27-beta.1 beta before the next release decision.

Release: https://github.com/openclaw/openclaw/releases/tag/v2026.5.27-beta.1

What changed in this beta

The beta includes a broad release-branch backport set with the highest-risk areas clustered around:

• Channel delivery and approvals: Discord, iMessage, Matrix, Slack, Telegram, Teams, QQBot, Google Chat, Feishu, and generic channel runtime metadata paths.
• Codex runtime behavior: watchdogs, shared app-server cleanup, native hook relay, dynamic tool quarantine reporting, workspace memory routing, and runtime model resolution.
• Gateway, sessions, and auth: session lock release on timeout abort, stale continuation handling, browser token expiry after auth rotation, Tailscale no-auth exposure rejection, device-role approval checks, and auth/profile routing fixes.
• Plugins, providers, and catalogs: approval action metadata, Plugin SDK approval command actions, lazy plugin runtime surface reflection, ClawHub display names, runtime catalog metadata, DeepInfra/OpenAI-compatible provider updates, retry-after parsing, and legacy memory provider migration.
• CLI, doctor, and status output: stricter malformed numeric/timeout option handling, actionable doctor restart guidance, unsupported tool schema warnings, and fast status preserving configured channel rows.
• Release/package validation: Docker runtime template checks, shrinkwrap/pin fixes, package inventory exclusions, beta smoke protections, and CI/release validation hardening.

Please focus testing on

• Upgrade from v2026.5.26 to v2026.5.27-beta.1, then run normal startup, openclaw status, and openclaw doctor.
• Send and approve tool actions from real channels, especially Discord, iMessage, Matrix, Slack, Telegram, Teams, Google Chat, Feishu, and QQBot.
• Exercise recovered or timeout-adjacent tool flows: file writes/edits that complete late, queued replies, late cleanup, final reply delivery, and fallback warnings.
• Run Codex sessions that use dynamic tools, workspace memory, native hooks, and model switching; watch for stuck runs, missing final replies, or incorrect tool quarantine messages.
• Test auth/profile routing: OpenAI-compatible providers, Codex OAuth, GitHub Copilot, provider-specific auth profiles, and configured model suffixes.
• Verify plugin/provider browsing and metadata: ClawHub/plugin display names, lazy runtime surfaces, runtime catalog markdown metadata, DeepInfra model browsing, and web-search provider config.
• Check security-sensitive boundaries: untrusted group prompts, Teams service URLs, unsafe Node runtime env overrides, side-effecting command wrappers, Discord guild actions, and device approval permissions.
• Validate package/runtime installs where possible, especially Docker runtime templates and bundled plugin startup after upgrade.

Feedback format

Please comment with:

• Version and install path tested.
• OS/runtime details.
• Channel or plugin under test.
• Exact command or user action.
• Expected result.
• Actual result.
• Logs or screenshots only if they do not contain secrets.

Known useful links

• Beta release: https://github.com/openclaw/openclaw/releases/tag/v2026.5.27-beta.1
• Previous stable release used for review context: https://github.com/openclaw/openclaw/releases/tag/v2026.5.26
• Compare context: https://github.com/openclaw/openclaw/compare/v2026.5.26...v2026.5.27-beta.1