Is OpenClaw v2026.6.11 stable?
-423/100Fact score: -423/100 · evidence-based, not a YES/NO verdict
“v2026.6.11 picked up fresh gated evidence: exec output can vanish, model/provider routing still trips on real configs, and companion surfaces are throwing punches. The score was already in the basement; now it found stairs.”
Stability Score
-423Base
80
Evidence
-423
Votes
-0
Survived
ignored
Curated
ignored
-423 evidence debt=-423
Affected components
acpapproval-boundaryauth-providerbrowsercodexcompactionconfig-migrationcrondesktop-macosdiscordemailexecgatewayhot-reloadlaunchagentmattermostmessage-deliverymodel-routingnode-surfaceplugin-installplugin-systemprompt-cacheprovider-authprovider-routingreasoningservice-envsession-statesessions-managementsignalskill-workshopslacksubagent-runtimetelegramtoolsuiupgradeusage-accountingweb-searchweb-toolswebchatwechatwecomwhatsappwindows-runtimeworkboard
Install this version
npm install openclaw@2026.6.11openclaw update --tag 2026.6.11Score Evidence
| Issue | Area | Type | Penalty | Reason |
|---|---|---|---|---|
| openclaw/openclaw issue 98416 → | sessions-management | kernel | -5 | Published release artifact lacks effective reply-session reentrancy guard, causing initialization conflicts. |
| openclaw/openclaw issue 98528 → | tools | kernel | -5 | Tool result transport regresses after first tool call per turn for exec/web_fetch/web_search. |
| openclaw/openclaw issue 98614 → | sessions-management | kernel | -5 | sessions_spawn fails with missing operator.write scope after upgrade from 2026.6.1. |
| openclaw/openclaw issue 98415 → | sessions-management | runtime | -4 | sessions_send can accept then silently fail on locked target session files. |
| openclaw/openclaw issue 98565 → | upgrade | runtime | -4 | Container image upgrades can skip upgrade migrations before gateway startup. |
| openclaw/openclaw issue 98419 → | config-migration | runtime | -4 | Plugin update path can rewrite projected config and trip size-drop protection. |
| openclaw/openclaw issue 98409 → | cron | runtime | -4 | Deleted recurring cron jobs can silently respawn on every gateway restart. |
| openclaw/openclaw issue 98379 → | approval-boundary | runtime | -3 | Control UI approval window can close while queued approvals remain pending. |
| openclaw/openclaw issue 98583 → | plugin-install | tool | -3 | @openclaw/acpx published workspace:* devDependency breaks npm gateway plugin install. |
| openclaw/openclaw issue 96878 → | web-search | tool | -3 | web_search providers and official plugin installation fail on the release train. |
| openclaw/openclaw issue 98557 → | telegram | channel-specific | -3 | Telegram rich delivery can leak standalone tool-call-shaped parameter wrappers. |
| openclaw/openclaw issue 98448 → | desktop-macos | runtime | -2 | Bonjour advertiser retries failed probing indefinitely and env-var disable is ineffective. |
| openclaw/openclaw issue 98232 → | launchagent | runtime | -2 | macOS LaunchAgent update can report handoff failure while replacement gateway is healthy. |
| openclaw/openclaw issue 97636 → | ui | runtime | -2 | Control UI can connect to the wrong gateway across base paths. |
| openclaw/openclaw issue 98392 → | approval-boundary | runtime | -2 | Control UI can show stale approval resolve errors on the wrong prompt. |
| openclaw/openclaw issue 98484 → | plugin-install | tool | -2 | ACPX plugin upgrade fails with blank/invalid npm-install path. |
| openclaw/openclaw issue 98573 → | windows-runtime | runtime | -2 | Gemini CLI backend fails on Windows because bare command misses .cmd handling. |
| openclaw/openclaw issue 98064 → | node-surface | runtime | -2 | iOS LAN pairing can drop bootstrap handoff token before chat reconnect. |
| openclaw/openclaw issue 97680 → | plugin-system | tool | -2 | Beta-tagged update can leave official external plugins on latest instead of requested beta tag. |
| openclaw/openclaw issue 98564 → | plugin-system | channel-specific | -1 | Mattermost repair installs external plugin but does not enable gateway startup. |
| openclaw/openclaw issue 98382 → | ui | other | -1 | Talk mode Chat WKWebView viewport can collapse to about 30px width on macOS. |
| openclaw/openclaw issue 98295 → | model-routing | tool | -1 | Custom Xiaomi MiMo models can generate invalid max_completion_tokens when maxTokens is omitted. |
| openclaw/openclaw issue 98540 → | ui | other | -1 | Composer can show idle state while an agent is still executing tools. |
| openclaw/openclaw issue 98566 → | ui | other | -1 | Chat workspace rail sections lack scrollbars and overflow long file lists. |
| openclaw/openclaw issue 97021 → | telegram | channel-specific | -1 | Telegram typing indicator can remain stuck in group forum/topic mode. |
| openclaw/openclaw issue 98282 → | telegram | channel-specific | -1 | Telegram forum-topic long streamed replies can be truncated in-place. |
| openclaw/openclaw issue 99061 → | channel-specific | -4 | WhatsApp outbound relay silently stops delivering replies mid-session on 2026.6.11 while model turns complete | |
| openclaw/openclaw issue 99068 → | discord | channel-specific | -2 | Discord replyToMode first repeats the native reply reference on every chunk, causing notification spam on 2026.6.11 |
| openclaw/openclaw issue 99060 → | subagent-runtime | kernel | -4 | Webchat orchestrator can deadlock after subagent completion events trigger premature NO_REPLY on 2026.6.11 |
| openclaw/openclaw issue 98842 → | plugin-system | runtime | -3 | Externalized plugins lose non-gateway public artifacts on packaged 2026.6.11 installs |
| openclaw/openclaw issue 98799 → | config-migration | runtime | -4 | install.sh config set channels --json overwrites existing channel configs without --merge on 2026.6.11 |
| openclaw/openclaw issue 98740 → | mattermost | channel-specific | -3 | Mattermost native slash commands return 401 after 2026.6.11 plugin externalization despite clean re-registration |
| openclaw/openclaw issue 98702 → | auth-provider | runtime | -3 | Inherited OpenAI OAuth is rejected for built-in openclaw runtime on 2026.6.11 openai-chatgpt-responses transport |
| openclaw/openclaw issue 98673 → | tools | kernel | -4 | 2026.6.11 can convert text tool results into image blocks, poisoning readable session history |
| openclaw/openclaw issue 98650 → | codex | kernel | -4 | Codex harness native hook relay unavailable causes every tool to fail-close on 2026.6.11 isolated multi-agent gateways |
| openclaw/openclaw issue 98588 → | auth-provider | runtime | -3 | anthropic/claude-opus-4-8 rejects tools.profile=coding schemas on 2026.6.11 while adjacent combos work |
| openclaw/openclaw issue 98532 → | provider-auth | kernel | -4 | Anthropic replay_invalid thinking block errors are not retried, leaving 2026.6.11 sessions permanently broken until /new |
| openclaw/openclaw issue 98426 → | compaction | kernel | -4 | Fallback to smaller-context models can irreversibly compact session context on 2026.6.11, silently losing data |
| openclaw/openclaw issue 99499 → | message-delivery | kernel | -5 | Completed local-provider/subagent work can remain stuck in pendingFinalDelivery and not reach the UI until a later message flushes it. |
| openclaw/openclaw issue 99346 → | webchat | kernel | -5 | WebChat direct restart recovery can route new turns into agent:main:boot while the main session remains conflicted. |
| openclaw/openclaw issue 99411 → | tools | kernel | -4 | Repeated exec/read output can disappear after aggregate tool-result truncation in the published release. |
| openclaw/openclaw issue 99522 → | desktop-macos | runtime | -4 | macOS app can mint a fresh device identity on every gateway reconnect because the app-group identity store is unusable. |
| openclaw/openclaw issue 99260 → | plugin-system | runtime | -4 | Official ACPX plugin cannot open keyed state because external trusted plugins are blocked by the release trust gate. |
| openclaw/openclaw issue 99387 → | provider-auth | runtime | -3 | Anthropic/Telegram schema rejection remains plausible on the release due to older signed-thinking replay/recovery behavior. |
| openclaw/openclaw issue 99502 → | windows-runtime | runtime | -2 | Windows gateway start can use a visible Scheduled Task terminal window despite the intended hidden launcher path. |
| openclaw/openclaw issue 99286 → | ui | runtime | -2 | macOS dashboard session deletion can do nothing because the embedded Web UI confirm dialog is not implemented. |
| openclaw/openclaw issue 99240 → | model-routing | tool | -2 | GitHub Copilot Claude model metadata can collapse /thinking options to off-only in the release. |
| openclaw/openclaw issue 99239 → | model-routing | tool | -1 | Bundled Anthropic model catalog lacks the current Sonnet entry, confusing model discovery and routing diagnostics. |
| openclaw/openclaw issue 99163 → | tools | runtime | -3 | Large chat.send attachment parse/stage payloads can hit Maximum call stack exceeded on the release. |
| openclaw/openclaw issue 99117 → | telegram | channel-specific | -1 | Telegram prompt context can duplicate assistant messages from session transcript and channel cache. |
| openclaw/openclaw issue 98753 → | cron | runtime | -3 | CLI health and cron list commands can close the gateway WebSocket with 1006 after doctor repair. |
| openclaw/openclaw issue 98685 → | sessions-management | runtime | -2 | Diagnostic queueDepth can stay positive after queued embedded messages are accepted and the active run ends. |
| openclaw/openclaw issue 98674 → | desktop-macos | ui | -1 | macOS app onboarding/install control can be unclickable for the reported release environment. |
| openclaw/openclaw issue 99951 → | codex | kernel | -5 | Codex provider route can run stale vendored @openai/codex in v2026.6.11 even when the host Codex CLI is current. |
| openclaw/openclaw issue 99704 → | channel-specific | -4 | WeChat/openclaw-weixin messages can be lost or duplicated and WebChat can miss WeChat sessions on v2026.6.11. | |
| openclaw/openclaw issue 99881 → | tools | runtime | -3 | After an image upload to a non-multimodal model, all later tool outputs can render as '(see attached image)' on v2026.6.11. |
| openclaw/openclaw issue 99603 → | hot-reload | kernel | -5 | Git-update hot reload can wipe dist/ mid-rebuild and crash-loop the gateway on v2026.6.11 production checkouts. |
| openclaw/openclaw issue 99550 → | codex | runtime | -3 | Codex OAuth gpt-5.5 turns can wait about 118s on failed responses_websocket retries before HTTP fallback succeeds. |
| openclaw/openclaw issue 98618 → | upgrade | runtime | -2 | After update to 2026.6.11, CLI/version metadata can still report 2026.6.10 and confuse doctor/plugin API checks. |
| openclaw/openclaw issue 100324 → | channel-specific | -4 | Official ClawHub WhatsApp plugin cannot start linked accounts on v2026.6.11 due to openKeyedStore trust checks. | |
| openclaw/openclaw issue 100066 → | provider-routing | kernel | -5 | Provider model registration error gives a validator-rejected config fix, causing gateway crash-loop risk on v2026.6.11. |
| openclaw/openclaw issue 100067 → | auth-provider | runtime | -5 | 2026.6.x auth-store migration can strand existing JSON OAuth/token profiles after upgrade to v2026.6.11. |
| openclaw/openclaw issue 100280 → | node-surface | runtime | -2 | Mobile pairing can advertise an unreachable LAN URL instead of usable Tailnet routes on v2026.6.11. |
| openclaw/openclaw issue 99964 → | sessions-management | runtime | -3 | Main sessions can roll over into .jsonl.reset when transcript mtime is newer than registry updatedAt on v2026.6.11. |
| openclaw/openclaw issue 95515 → | runtime | -4 | Email channel config migration can still write invalid groupAllowFrom during upgrades; latest v2026.6.11 keeps the permissive path. | |
| openclaw/openclaw issue 100458 → | skill-workshop | tool | -3 | Skill and core read/output tools can truncate or lose longer results, and Skill Workshop proposal state can disappear on v2026.6.11. |
| openclaw/openclaw issue 100471 → | cron | runtime | -4 | agentTurn cron jobs can leave running_at_ms uncleared after completion, causing later scheduled runs to be skipped on v2026.6.11. |
| openclaw/openclaw issue 100427 → | approval-boundary | kernel | -5 | Policy plugins cannot see resolved served-model identity at any blocking hook, creating a failed-open confidentiality enforcement gap observed on v2026.6.11. |
| openclaw/openclaw issue 100312 → | compaction | kernel | -5 | Compaction-timeout recovery can still leave assistant-last session state on v2026.6.11, breaking continuation and misattributing infrastructure errors to model fallback. |
| openclaw/openclaw issue 100566 → | provider-routing | runtime | -4 | thought_signature 400 regression breaks reasoning-model requests on v2026.6.11. |
| openclaw/openclaw issue 100677 → | compaction | kernel | -5 | Late compaction count reconcile throws TypeError repeatedly on v2026.6.11. |
| openclaw/openclaw issue 100783 → | exec | runtime | -4 | exec tool output buffer accumulates across commands, causing truncation or hangs on Linux v2026.6.11. |
| openclaw/openclaw issue 100787 → | webchat | channel-specific | -3 | WebChat renders tool call results as screenshots/images, making output unreadable on v2026.6.11. |
| openclaw/openclaw issue 100941 → | gateway | runtime | -4 | Concurrent in-process tool-to-gateway WebSocket fan-out drops connections with 1006 on v2026.6.11. |
| openclaw/openclaw issue 100944 → | message-delivery | channel-specific | -3 | Signal DMs can be silently dropped on reply-session initialization conflict in v2026.6.11. |
| openclaw/openclaw issue 100988 → | subagent-runtime | runtime | -4 | sessions_spawn from a CLI-harness parent can clamp children to bridge tools, losing exec/read/write on v2026.6.11. |
| openclaw/openclaw issue 101049 → | plugin-install | runtime | -3 | Microsoft Teams plugin install can leave an empty dependency tree while appearing loaded on v2026.6.11. |
| openclaw/openclaw issue 101459 → | model-routing | runtime | -3 | openai-responses sends system instead of developer role for reasoning models on v2026.6.11. |
| openclaw/openclaw issue 101467 → | webchat | channel-specific | -3 | TUI/WebChat tool output renders as images after v2026.6.11 upgrade. |
| openclaw/openclaw issue 101537 → | cron | runtime | -4 | Scheduled cron wake can be skipped as disabled, leaving the main agent unwoken on v2026.6.11. |
| openclaw/openclaw issue 101570 → | model-routing | runtime | -3 | Thinking content can leak into text replies with severe repetition on v2026.6.11 model paths. |
| openclaw/openclaw issue 101672 → | provider-routing | runtime | -4 | Session model/runtime binding drift persists across doctor repair because provider ownership misses literal openai id. |
| openclaw/openclaw issue 101673 → | plugin-system | runtime | -3 | Plugin-registry workspace caching bug can expose stale plugin state on v2026.6.11. |
| openclaw/openclaw issue 101674 → | reasoning | tool | -3 | Sonnet 5 thinking.type.enabled is rejected and silently retried with thinking off on v2026.6.11. |
| openclaw/openclaw issue 101675 → | sessions-management | runtime | -4 | Interrupted sessions can become unrecoverable after gateway restart because transcript tail is not resumable. |
| openclaw/openclaw issue 101676 → | sessions-management | runtime | -3 | Live model switch mid-attempt terminates the run instead of retrying against the new model. |
| openclaw/openclaw issue 101814 → | message-delivery | kernel | -5 | All channels can enter a one-message-then-silence state on 2026.6.11, with duplicate replies and restart/session cleanup recovery. |
| openclaw/openclaw issue 101850 → | session-state | kernel | -5 | Long-session missing tool-result race remains reproducible on 2026.6.11, producing synthetic errors and lost tool outputs. |
| openclaw/openclaw issue 102006 → | exec | kernel | -5 | Exec/tool calls can wedge after aborts or rapid gateway-touching operations on 2026.6.11, requiring gateway restart or session surgery. |
| openclaw/openclaw issue 102290 → | desktop-macos | runtime | -3 | Official 2026.6.11 macOS DMG verifies but installed app fails codesign/spctl on Apple Silicon. |
| openclaw/openclaw issue 102401 → | wecom | channel-specific | -3 | WeCom group messages can wedge the session when consecutive instructions arrive on 2026.6.11. |
| openclaw/openclaw issue 102580 → | telegram | kernel | -5 | Documented Telegram chunking keys are rejected by config validation, causing gateway crash loop on restart. |
| openclaw/openclaw issue 102295 → | slack | channel-specific | -1 | Slack allowBots with wildcard allowFrom silently drops bot-authored messages on 2026.6.11. |
| openclaw/openclaw issue 102578 → | telegram | channel-specific | -3 | Telegram DM block-streaming can silently drop the tail of long replies on 2026.6.11. |
| openclaw/openclaw issue 103137 → | browser | tool-specific | -3 | Browser evaluate aborts around 15s on Windows/2026.6.11 and ignores the requested timeout. |
| openclaw/openclaw issue 103161 → | provider-routing | runtime | -2 | LLM provider rejection details are not persisted to disk on 2026.6.11, leaving failure diagnosis blind. |
| openclaw/openclaw issue 103198 → | webchat | tool-specific | -3 | WebChat image attachments pass placeholder IDs to image tooling instead of media-store file paths on 2026.6.11. |
| openclaw/openclaw issue 103342 → | auth-provider | runtime | -4 | Control UI token-auth behind Nginx can enter an infinite redirect / forced logout loop on 2026.6.11. |
| openclaw/openclaw issue 103385 → | web-tools | tool-specific | -2 | Google Meet browser transport fails silently on non-English Meet UI labels with the 2026.6.11 plugin. |
| openclaw/openclaw issue 103520 → | model-routing | kernel | -5 | Configured disallowed models can be silently substituted with the first allowed catalog entry, creating unrequested paid-provider usage on 2026.6.11. |
| openclaw/openclaw issue 104480 → | message-delivery | channel-specific | -2 | Large Telegram media sends can exceed the CLI-to-gateway RPC deadline, reporting false failure after successful delivery and encouraging duplicate retries. |
| openclaw/openclaw issue 104389 → | exec | runtime | -3 | Auto-backgrounded exec failures can post raw internal system cards, agent ids, and command lines to end users. |
| openclaw/openclaw issue 103734 → | codex | runtime | -4 | Codex usage-limit promptError is surfaced without triggering configured model fallback on v2026.6.11. |
| openclaw/openclaw issue 104320 → | web-search | tool-specific | -3 | infer web search rejects the installed Exa plugin SecretRef target as unknown on v2026.6.11. |
| openclaw/openclaw issue 104050 → | workboard | tool-specific | -2 | workboard claim() is not idempotent for already-claimed dependency-backed cards and blocks repeat claims. |
| openclaw/openclaw issue 102175 → | prompt-cache | runtime | -3 | Embedded prompt cache can break across room-event, policy, and Responses boundaries, replaying full context instead of native continuation. |
| openclaw/openclaw issue 103802 → | acp | runtime | -4 | Gateway force-injects default thinking config that ACP backends do not advertise, causing every ACP spawn to fail. |
| openclaw/openclaw issue 103804 → | service-env | runtime | -3 | launchd service-env generation double-quotes values, breaking AWS_REGION-derived Bedrock hostnames. |
| openclaw/openclaw issue 103710 → | provider-routing | runtime | -3 | Moonshot/Kimi requests can ECONNRESET or timeout through the SSRF-guarded fetch path while direct undici/curl works. |
| openclaw/openclaw issue 103684 → | message-delivery | kernel | -5 | Final replies can never deliver because dispatcher.waitForIdle blocks forever on a beforeDeliver hook, leaving the lane blocked. |
| openclaw/openclaw issue 103636 → | gateway | kernel | -4 | Backend loopback relay from the host interface can lose operator scope, rejecting agent RPC with missing operator.write. |
| openclaw/openclaw issue 103386 → | web-tools | tool-specific | -2 | googlemeet leave can report success while the Chrome participant remains in the meeting. |
| openclaw/openclaw issue 99947 → | codex | runtime | -4 | Codex harness mirrored-session-history reads can fail for ephemeral sessions and app-server cleanup can retire shared apps mid-turn. |
| openclaw/openclaw issue 89278 → | provider-auth | runtime | -3 | Revoked OpenAI refresh tokens on v2026.6.11 can surface as generic 10s auth-refresh timeouts, hiding the actionable reauth path. |
| openclaw/openclaw issue 99470 → | session-state | runtime | -3 | Delivery-mirror transcript entries can leak into model prompts and duplicate prior user/assistant transcript content. |
| openclaw/openclaw issue 104763 → | exec | tool-specific | -3 | Exec output longer than a few lines can silently disappear instead of reaching the user as text. |
| openclaw/openclaw issue 104843 → | webchat | channel-specific | -1 | WebChat transcript can jump to the top while streaming, especially with TTS enabled. |
| openclaw/openclaw issue 104721 → | tools | kernel | -5 | Tool results can collapse into a literal “see attached image” placeholder across channels, making tool output unusable. |
| openclaw/openclaw issue 104538 → | launchagent | runtime | -4 | macOS launchd supervisor restart on config change can exit without relaunching, leaving the gateway dead until manual kickstart. |
| openclaw/openclaw issue 104783 → | provider-routing | runtime | -3 | Groq provider can send oversized requests for tiny prompts, hitting provider TPM limits. |
| openclaw/openclaw issue 105305 → | browser | tool-specific | -3 | Browser screenshots can poison subsequent tool-result rendering so later outputs are treated as image analysis. |
| openclaw/openclaw issue 102081 → | approval-boundary | kernel | -4 | Darwin exec allowlist matches still require explicit approval, making documented allowlist auto-execution unavailable after auto-update. |
| openclaw/openclaw issue 102270 → | plugin-install | runtime | -3 | Plugin installs can skip native binding builds on Node 24, leaving better-sqlite3 based plugins unable to load. |
| openclaw/openclaw issue 101335 → | telegram | channel-specific | -2 | Telegram isolated ingress serializes same-chat bursts before they can coalesce, creating separate turns instead of one collected follow-up. |
| openclaw/openclaw issue 102020 → | signal | channel-specific | -3 | Signal sessions can fail on the second message with reply-session initialization conflicts. |
| openclaw/openclaw issue 105664 → | exec | kernel | -5 | exec output can disappear from agent-visible context on 2026.6.11, with process polling returning no retained session. |
| openclaw/openclaw issue 105629 → | desktop-macos | runtime | -3 | Apple Watch companion build tied to the 2026.6.11 iOS release crashes immediately on launch on physical hardware. |
| openclaw/openclaw issue 106374 → | provider-routing | runtime | -3 | openai/gpt-5.6-terra is listed/configurable but Gateway rejects runs because the model registry entry is missing. |
| openclaw/openclaw issue 106203 → | node-surface | runtime | -3 | Healthy remote Windows nodes expose system tools but no node_exec surface in Codex/WebChat on 2026.6.11. |
| openclaw/openclaw issue 106277 → | codex | tool-specific | -3 | codex_app__automation_update exposes a null schema type, causing Codex endpoint HTTP 400 on 2026.6.11. |
| openclaw/openclaw issue 106080 → | usage-accounting | runtime | -2 | Model cost accounting reports $0 because model_capability_cache never populates, reproduced on 2026.6.11. |
| openclaw/openclaw issue 105528 → | tools | runtime | -4 | exec/read tools can silently return empty output on Windows across the 2026.6.x release line. |