# ClawStat.us — OpenClaw v2026.6.6

> Should you update to OpenClaw v2026.6.6? Verdict: 🔄 Wait for next release (medium confidence). Assessed 2026-06-14.

Multiple critical regressions plague v2026.6.6; pre-release v2026.6.8-beta.1 contains fixes — wait for stable release.

## Verdict

- Subject: OpenClaw v2026.6.6
- Recommendation: 🔄 Wait for next release
- Confidence: medium
- Assessed at: 2026-06-14T18:30:01.786540+00:00
- Latest release: v2026.6.6 (published 2026-06-12)
- Latest pre-release: v2026.6.8-beta.1
- npm: 2026.6.6

## Why this verdict

v2026.6.6 shipped with valuable security hardening, Telegram delivery improvements, iMessage recovery, browser/MCP connectivity, Control UI performance, and provider support. However, it also introduced several critical regressions that affect core functionality: memory_search intermittently fails (#90361), containerized deploys cannot use externalized channel plugins (#92516, #92452), and the agent session fails to start after gateway restart (#92511). High-severity issues include DeepSeek prompt cache failure (#91016), isolated cron failures (#91363), WeChat duplicate sessions (#92384), model fallback notice leaks (#92473), Feishu DM dispatch failures (#92364), and isolated cron delivering internal debug text to Telegram (#92535). The previous assessment (2026-06-14) recommended waiting for the next release, and that recommendation remains valid. A pre-release v2026.6.8-beta.1 was published on 2026-06-13 with fixes pending for the next stable release. No material change in evidence has occurred since the previous assessment; the same critical regressions persist, and the pre-release path to resolution is unchanged. Users who rely on memory_search, containerized deployments, or agent session reliability should avoid this version. Those who need the security hardening and Telegram improvements can consider upgrading with precautions, but the safer path is to wait for the stable release that incorporates the pre-release fixes.

## Reasons to update

- Security hardening across transcripts, sandbox binds, host environment, MCP stdio, Codex HTTP access, and more (#91529, #91618, #91615, #91619, #91741, #91745, #91746, #91748, #91749, #91750, #91751, #91752, #91763, #89938)
- Telegram delivery is safer and more coherent with account-scoped topics, streamed text survival, /compact support, and dedupe improvements (#91189, #88682, #89588, #90212, #91876, #91874, #91904, #91478, #91915)
- iMessage recovery and delivery hardened with always-on inbound restart, durable echo markers, and idle approval discovery (#91335, #91449, #88969, #88530, #91783, #91785)
- Browser and MCP connectivity gains existing-session CDP support, WebSocket validation, and Streamable HTTP loopback transport (#91422, #89851, #91736, #91747, #91451, #80143)
- Control UI startup and first-reply latency reduced via cached model metadata and lazy slash-command loading (#91531, #91538, #91568, #91583, #91598)
- Provider support expands with OpenRouter OAuth, Claude Fable 5 adaptive thinking, and Gemma 4 reasoning preservation (#91830, #91882, #91590, #88630, #88768, #91696)

## Reasons to hold off

- Critical regression: memory_search intermittently returns 'index metadata is missing' despite valid index (#90361, affects this version, Clawsweeper: keep_open, no fix yet)
- Critical regression: containerized deploys cannot use externalized channel plugins due to openKeyedStore trust gate (#92516, affects this version, Clawsweeper: keep_open, no fix yet)
- Critical regression: msteams ClawHub install fails with same trust gate issue (#92452, affects this version, Clawsweeper: keep_open, no fix yet)
- Critical regression: agent session doesn't start after gateway restart, requires second restart (#92511, affects this version, Clawsweeper: keep_open, no fix yet)
- High regression: DeepSeek Prompt Cache completely fails, costing ~$6/hour (#91016, affects this version, Clawsweeper: keep_open, no fix yet)
- High regression: isolated cron consistently fails with 'LLM request failed' (#91363, affects this version, Clawsweeper: keep_open, no fix yet)
- High regression: WeChat channel spawns duplicate direct+group sessions (#92384, affects this version, Clawsweeper: keep_open, no fix yet)
- High regression: model fallback notice leaks to external user-facing channels (#92473, affects this version, Clawsweeper: keep_open, no fix yet)
- High regression: Feishu DM dispatch fails with replies=0 (#92364, affects this version, Clawsweeper: keep_open, no fix yet)
- High regression: isolated cron can deliver unavailable-tool self-debug text to Telegram (#92535, affects this version, Clawsweeper: keep_open, no fix yet)
- High regression: tasks audit produces 154 false findings on healthy system (#92401, affects this version, Clawsweeper: keep_open, no fix yet)
- High regression: openai-responses adapter sends system prompt in input instead of instructions (#92400, affects this version, Clawsweeper: keep_open, no fix yet)

## Context

- Ongoing major issues not specific to this version include Codex-backed Telegram timeouts (#87744), stale worker process accumulation (#76171), gateway heap growth (#89315), Control UI Raw mode disabled (#59330), and others — these exist regardless of version and should not drive the verdict.

## Known issues (24)

- #90361 [critical · regression · unfixed] [Bug]:Intermittent memory_search "index metadata is missing" despite valid builtin memory index; likely search/reindex race. Locally hotfixed. — 3 👍
- #92516 [critical · regression · unfixed] Containerized/self-hosted deploys can't use externalized channel plugins: openKeyedStore is gated to trusted plugins, with no supported way to trust a self-hosted channel — 1 👍
- #92452 [critical · regression · unfixed] msteams: ClawHub install can never pass the 6.x keyed-store trust gate (official catalog is npmSpec-only) — channel crash-loops with no actionable error — 1 👍
- #92511 [critical · regression · unfixed] Agent session doesn't start after gateway restart (requires second restart) — 1 👍
- #91016 [high · active · open] ⚠️ 升级 2026.6.1 后 DeepSeek Prompt Cache 完全失效，一小时烧掉约 $6 — 5 👍
- #91363 [high · active · open] Isolated cron consistently fails with "LLM request failed" on model-call-started phase — 4 👍
- #92384 [high · regression · unfixed] [Bug]: WeChat channel: same DM peer spawns duplicate direct+group sessions — outbound target classifier defaults bare IDs to "group" — 1 👍
- #92473 [high · regression · unfixed] Model fallback notice leaks to external user-facing channels — 1 👍
- #92364 [high · regression · unfixed] [Bug] Feishu DM dispatch complete (replies=0) — still present in v2026.6.5, multiple past reports — 1 👍
- #92535 [high · regression · unfixed] [Bug]: isolated cron can deliver unavailable-tool self-debug to Telegram — 1 👍
- #92401 [high · regression · unfixed] tasks audit: 154 findings on a healthy system — false 'dead agent' signals for long-running subagents + timestamp noise flood — 1 👍
- #92400 [high · regression · unfixed] [Bug]: `openai-responses` adapter sends system prompt in `input` instead of `instructions` — 1 👍
- #88312 [critical · regression · unfixed] [Bug]: [Regression] 2026.5.27: Codex app-server turn-completion stall returns ("Codex stopped before confirming the turn was complete") — regression of #84076 (fixed by #85107) — 4 👍
- #86508 [critical · regression · unfixed] [Bug]: EmbeddedAttemptSessionTakeoverError during Discord runs: session file changed while embedded prompt lock was released — 4 👍
- #53599 [critical · regression · unfixed] [Bug]: Chrome extension browser relay removed with no cross-machine  replacement (regression) — 5 👍
- #38327 [critical · regression · unfixed] [Bug] "Cannot convert undefined or null to object" in 2026.3.2 with google-vertex/gemini-3.1-pro-preview — 3 👍
- #92361 [medium · regression · unfixed] [Bug]:  Tool availability evaluator silently ignores empty `allOf`/`anyOf` groups during expression normalization — 1 👍
- #92338 [medium · regression · unfixed] [Bug]: cron-isolated agent runs never emit model.usage diagnostic events (OTel blind to all scheduled spend) — 1 👍
- #32473 [high · regression · unfixed] [Bug]: control ui requires device identity (use HTTPS or localhost secure context) — 5 👍
- #92523 [low · regression · unfixed] Bug: Orphaned TaskFlows in `waiting` status permanently block agent heartbeats (requests-in-flight deadlock) — 1 👍
- #92506 [low · regression · unfixed] [Bug]: /status usage missing for OpenAI Codex synthetic auth after 2026.6.6 — 1 👍
- #92330 [low · regression · unfixed] exec host=node always returns SYSTEM_RUN_DENIED when exec-approvals.json has a socket token (even in YOLO mode) — 1 👍
- #92327 [low · regression · unfixed] [v2026.6.5] WebChat UI: sessions_send inter-session messages render alongside assistant messages without visual separation (regression of #89191) — 1 👍
- #92423 [low · regression · unfixed] [Docs/Question]: Recommended pattern for telemetry-only message lifecycle capture — 1 👍

## What's new

### Features

- CLI progress: emit Claude CLI commentary progress events and bridge inter-tool commentary into channel progress without exposing internal protocol scaffolding (#89834, #90883)
- Observability: allow trusted diagnostics channels to capture tool input/output content, add first-assistant-event traces, and warn on slow initial replies (#91256, #91568, #91583)
- Plugins/ClawHub: dogfood reusable package publishing, let dry runs skip publish approval, allow declared installed trusted hooks, report managed plugin version drift

### Fixes

- Security hardening across transcripts, sandbox binds, host environment, MCP stdio, Codex HTTP access, native search policy, elevated sender checks, deleted-agent ACP bypasses, loopback tools, Discord moderation, and Teams group actions; exec approvals now fail closed on timeout (#91529, #91618, #91615, #91619, #91741, #91745, #91746, #91748, #91749, #91750, #91751, #91752, #91763, #89938)
- Telegram delivery: account-scoped topics, streamed text survives tool calls, /compact works on generic ingress, callback handling uses concrete APIs, draft chunking shared, durable dispatch dedupe moved into SDK, unauthorized DM text excluded from cache and prompt context (#91189, #88682, #89588, #90212, #91876, #91874, #91904, #91478, #91915)
- iMessage recovery and delivery: always-on inbound restart, durable echo markers, block streaming, idle approval discovery, hardened outbound transport, actionable inbound startup diagnostics (#91335, #91449, #88969, #88530, #91783, #91785)
- Browser and MCP connectivity: existing-session CDP support, WebSocket validation, default-profile cdpUrl handling, safer browser-output boundaries, Streamable HTTP loopback transport, corrected OAuth/SSE authorization handling, broader schema compatibility (#91422, #89851, #91736, #91747, #91451, #80143)
- Control UI startup and first-reply latency reduced via cached model metadata, removal of startup catalog wait, lazy slash-command loading, and first-event tracing with slow-reply diagnostics (#91531, #91538, #91568, #91583, #91598)
- Provider support: OpenRouter OAuth onboarding, Claude Fable 5 adaptive thinking, Codex sessions keep correct compaction ownership, local models skip guardian review, dynamic tool progress normalizes cleanly, Gemma 4 reasoning replay preserved (#91830, #91882, #91590, #88630, #88768, #91696)

## Platform impact

- windows: low
- macos: low
- linux: high
- discord: low
- slack: low
- telegram: medium

## Community sentiment

Community sentiment is mixed: the security hardening and Telegram improvements are welcomed, but frustration is high due to multiple critical regressions affecting core functionality (memory_search, container deploys, agent session start). Several issues have garnered multiple 👍 reactions and comments, indicating widespread impact. The pre-release v2026.6.8-beta.1 is seen as a necessary fix, and users are advised to wait.

---
Machine-readable JSON: https://clawstat.us/latest.json · RSS: https://clawstat.us/feed.xml · Human page: https://clawstat.us/
Generated automatically by ClawStat.us — every field is an automated assessment; verify against the linked GitHub issues.