[v2026.6.5] [OpenClaw] Is it stable?

[github-actions]

Stability Score Evidence — v2026.6.5

The release has crossed the 24h line and found fresh cron, provider, hook, iMessage, heartbeat, and image-tool traps. The 3am deploy goblin brought receipts. Operator audit on 2026-06-13 strict-applied the new evidence-scope rule (only explicit v2026.6.5 stable still-affects/reproduced proof), trimming beta-only and migration-candidate evidence.

Stability adjustments

Affected components

• codex
• compaction
• cron
• cron-runtime
• discord
• feishu
• heartbeat
• image-tool
• imessage
• local-extensions
• memory
• memory-wiki
• message-delivery
• model-routing
• node-surface
• opencode
• plugin-install
• provider-integration
• reasoning
• runtime
• sessions-management
• slack
• update-system
• upgrade
• webchat

Evidence

Issue	Area	Type	Severity	Penalty	Reason
openclaw/openclaw#91804	reasoning	runtime	5	-5	Internal reasoning leakage exposed to users
openclaw/openclaw#91873	slack	channel-specific	2	-2	Slack channel silently drops after upgrade due plugins.allow regression
openclaw/openclaw#91823	update-system	runtime	3	-3	Linux systemd updater still runs inside Gateway process tree
openclaw/openclaw#91815	provider-integration	tool-specific	1	-1	SecretRef-backed Google Chat auth health mismatch in CLI commands
openclaw/openclaw#91772	plugin-install	tool-specific	3	-3	Plugin install fails with npm EOVERRIDE from shipped workspace overrides
openclaw/openclaw#91902	memory	runtime	5	-5	memory index --force writes incompatible meta providerKey and breaks memory_search
openclaw/openclaw#91766	cron	tool-specific	1	-1	cron status reports legacy storePath
openclaw/openclaw#91872	node-surface	channel-specific	2	-2	Android node mode chat.send never reaches Gateway v2026.6.5
openclaw/openclaw#91865	runtime	runtime	2	-2	OTLP log records are content-free and uncorrelatable through stable 2026.6.5
openclaw/openclaw#91816	upgrade	runtime	4	-4	Updater to 2026.6.5 fails from beta with missing SDK export; Feishu stops replying
openclaw/openclaw#91806	model-routing	runtime	4	-4	UI shows OpenAI but turns execute via DeepSeek and history forks
openclaw/openclaw#91861	message-delivery	runtime	2	-2	MEDIA directive replies duplicate text through channel.message split path
openclaw/openclaw#88332	codex	tool-specific	1	-1	Codex native status checks surface as channel-visible failures
openclaw/openclaw#91854	memory	tool-specific	1	-1	Dreaming narrative lane gets apply_patch and emits recurring error noise
openclaw/openclaw#91892	cron-runtime	runtime	5	-5	Cron jobs stall during AI model calls
openclaw/openclaw#91899	feishu	channel-specific	2	-2	Feishu WebSocket reconnect gives up after one retry
openclaw/openclaw#91778	memory	runtime	5	-5	memory_search index metadata missing since 2026.6.1 and reproduced on 2026.6.5
openclaw/openclaw#91756	memory-wiki	runtime	3	-3	memory-wiki bridge artifact visibility still unresolved in v2026.6.5
openclaw/openclaw#91839	sessions-management	runtime	5	-5	Terminal provider model_not_available can trigger chat.history storm and event-loop starvation
openclaw/openclaw#90023	memory	runtime	4	-4	QMD memory_search zero-hit queries trigger synchronous force sync and stall turns
openclaw/openclaw#91431	discord	channel-specific	1	-1	/export-session fails silently on Discord/channel-backed setup
openclaw/openclaw#91821	memory	tool-specific	1	-1	memory search --json prints results but does not exit with QMD backend
openclaw/openclaw#88929	feishu	channel-specific	2	-2	Feishu streaming card typewriter/final-content truncation regression
openclaw/openclaw#74586	memory	runtime	4	-4	Embedded run aborts memory_search tool calls and misclassifies timeout
openclaw/openclaw#89766	cron-runtime	runtime	5	-5	Isolated cron lanes leak on claude-cli backend until restart
openclaw/openclaw#91009	codex	tool-specific	5	-5	Codex PreToolUse hook relay spawns CPU-bound processes and stalls gateway RPC
openclaw/openclaw#86538	sessions-management	runtime	4	-4	Session write-lock timeouts block subagent delivery lanes
openclaw/openclaw#91971	plugin-install	tool-specific	1	-1	Documented @openclaw/llama-cpp-provider cannot be installed from npm or ClawHub on 2026.6.5
openclaw/openclaw#92058	cron-runtime	runtime	3	-3	failureAlert.after=1 never fires; errored cron runs stay delivery_status=not-requested on 2026.6.5
openclaw/openclaw#91944	cron-runtime	runtime	4	-4	Cron doctor preflight can restore stale legacy JSON schedules after 2026.6.5 upgrade
openclaw/openclaw#91920	cron-runtime	runtime	3	-3	Cron jobs fail on 2026.6.5 with fallback-summary/provider failure cascade
openclaw/openclaw#92097	provider-integration	runtime	4	-4	SecretRef-managed custom providers fail auth with 503 after update to 2026.6.5
openclaw/openclaw#92009	model-routing	runtime	3	-3	Configured google/gemini-3.1-pro-preview default model cannot be inspected or executed in 2026.6.5
openclaw/openclaw#91918	local-extensions	kernel	5	-5	before_tool_call/after_tool_call hooks register but never fire on 2026.6.5, bypassing safety hooks
openclaw/openclaw#91930	opencode	tool-specific	1	-1	OpenCode provider model catalog misses live Zen models on OpenClaw 2026.6.5
openclaw/openclaw#91916	feishu	channel-specific	2	-2	Feishu card streaming truncates long responses and can overwrite correct replies on 2026.6.5
openclaw/openclaw#92148	provider-integration	runtime	4	-4	DeepSeek custom OpenAI-compatible provider misroutes to OpenAI Responses API and fails with 401 on 2026.6.5
openclaw/openclaw#92142	imessage	channel-specific	2	-2	iMessage final replies fail under SIP because reply_to is passed despite reply actions disabled on 2026.6.5
openclaw/openclaw#92038	heartbeat	kernel	5	-5	NO_REPLY heartbeat silence is counted as empty output, cascading retries and fallback failures on 2026.6.5
openclaw/openclaw#92104	image-tool	tool-specific	2	-2	Image/PDF tools report Unknown model for every provider on 2026.6.5 while agent models work
openclaw/openclaw#92366	upgrade	runtime	5	-5	Upgrade from 2026.5.27 to 2026.6.5 causes broken service/config path and rollback
openclaw/openclaw#92344	compaction	kernel	5	-5	npm latest 2026.6.5 lacks the /compact reply delivery fix from #90212
openclaw/openclaw#92189	sessions-management	runtime	4	-4	sessions_spawn can zero target agent tools by intersecting requester deny rules on 2026.6.5
openclaw/openclaw#92182	slack	channel-specific	2	-2	Slack account-level streaming.mode=off is ignored and partial drafts post as new messages on 2026.6.5
openclaw/openclaw#92327	webchat	channel-specific	1	-1	WebChat sessions_send inter-session messages render without visual separation on v2026.6.5
openclaw/openclaw#92406	upgrade	runtime	3	-3	Legacy streamMode field can block 3.x-to-6.x config upgrade on 2026.6.5
openclaw/openclaw#92242	message-delivery	runtime	4	-4	Delivery bundle emits message_sent hooks but never records sent messages on 2026.6.5
openclaw/openclaw#92201	provider-integration	runtime	4	-4	Anthropic thinking signatures can be invalid on replay and recovery does not catch the genericized error on 2026.6.5

Stats

• Evidence penalty: -149
• Vote penalty: website-calculated
• Score formula: website-calculated from data/settings.json

Operator refresh notes — 2026-06-11

• Included in this refresh: image/pdf tool: 'Unknown model' for every provider, while the same models work as agent models (2026.5.28 and 2026.6.5) openclaw/openclaw#92104 — gated to v2026.6.5.

Operator refresh notes — 2026-06-12

• Included in this refresh: [Bug]: OpenClaw Upgrade Bug Report: 2026.5.27 → 2026.6.5 openclaw/openclaw#92366, [Bug]: npm latest 2026.6.5 still lacks #90212 /compact reply delivery fix openclaw/openclaw#92344, [Bug] sessions_spawn zeroes target agent tools by intersecting requester's effective deny (2026.6.5 regression) openclaw/openclaw#92189, [Bug]: 2026.6.5 ignores Slack account-level streaming.mode=off; partial draft snapshots posted as new messages instead of coalesced edits openclaw/openclaw#92182, [v2026.6.5] WebChat UI: sessions_send inter-session messages render alongside assistant messages without visual separation (regression of #89191) openclaw/openclaw#92327, [Bug]: streamMode field in 3.x config blocks upgrade to 6.x (schema rejects unknown field) openclaw/openclaw#92406, [regression] delivery bundle emits message_sent hooks but never calls recordSentMessage openclaw/openclaw#92242, Embedded runner: freshly streamed thinking signatures intermittently invalid on replay (Anthropic); recovery wrapper never fires because error text is genericized openclaw/openclaw#92201.
• All added candidates passed deterministic v2026.6.5 gate; score bucket remains score:0-9.

Operator refresh notes — 2026-06-13 (strict scope audit)

Applied the updated OPERATOR_POD.md §2.1 Evidence scope default rule. Removed 17 carry-forward / fixed / beta-only items that lacked explicit v2026.6.5 stable still-affects proof. Kept 48 items with explicit v2026.6.5 stable evidence (title, version field, body, or maintainer comment).

• Removed (migration candidates → migrate to their own version issues if needed):
  • [Bug]: Upgrade leaves repeated shared SQLite plugin install metadata conflict warnings for codex/discord openclaw/openclaw#90418 — beta-only: only mentions 2026.6.5-beta.2; no stable proof
  • MCP remote OAuth: two bugs block streamable-http servers (regex \: under /u; OAuth errors surfaced as [object Response]) openclaw/openclaw#91433 — beta-only: 'Reproduced on 2026.6.5-beta.2'
  • Regression: Feishu card delivery loses tool-call progress after v2026.5.19 (likely PR #83468) openclaw/openclaw#85439 — beta-only: 'OpenClaw main package: 2026.6.5-beta.2'; reporter asks ETA for stable fix
  • Bug: Internal thinking process exposed to user via Feishu openclaw/openclaw#88233 — beta-only: 'OpenClaw version: 2026.6.5-beta.2'
  • [Bug]: In-turn reasoning dropped on multi-turn tool replay for non-400 openai models (gemma4/vLLM) — silent agentic-quality regression openclaw/openclaw#91645 — beta-only: testing on 2026.6.5-beta.5 only
  • [Bug]: OpenAI Codex OAuth migration leaves stale default profile and codexPlugins app inventory can fail Codex harness openclaw/openclaw#91352 — beta-only: 'OpenClaw 2026.6.5-beta.2'
  • @openclaw/qqbot: missing approvalCapability + exec card omits command openclaw/openclaw#91890 — fixed/already-implemented: 'Current main and the v2026.6.5 release source already expose QQ Bot native approval'
  • opencode-go provider: qwen3.7-max, qwen3.7-plus, minimax-m3 missing from static model catalog → "Unknown model" (model_not_found) openclaw/openclaw#91238 — beta-only: '2026.6.5-beta.2 (beta)'
  • Managed restart shutdown can abort active main turns without restart-recovery state openclaw/openclaw#91355 — beta-only: 'Version observed: OpenClaw 2026.6.5-beta.2'
  • Current-session message-tool replies can be replaced by private bookkeeping finals openclaw/openclaw#91330 — beta-only: 'Version observed: OpenClaw 2026.6.5-beta.2'
  • [Feedback] Stop shipping features. Start shipping a product that works. openclaw/openclaw#90974 — beta-only feedback: only mentions 2026.6.5-beta.2 changelog
  • [Bug]: gateway opens an empty memory database when main.sqlite is absent during the index swap, leaving memory_search paused with "index metadata is missing" until restart openclaw/openclaw#91216 — beta-only: 'Live repro here on 2026.6.5-beta.1'
  • getSubagentRunsSnapshotForRead does an uncached full SQLite re-read on every call; many consumers drive it to ~22 Hz and pin a CPU core openclaw/openclaw#91517 — beta-only: 'In OpenClaw 2026.6.5-beta.2'
  • [Core bug] /compact reply silently dropped (replies=0) on all channels in 2026.6.1 openclaw/openclaw#90185 — duplicate of #92344 which directly tracks /compact reply on stable 2026.6.5
  • Codex Apps connector calls bypass pre-execution before_tool_call / PreToolUse hooks openclaw/openclaw#91454 — beta-only: 'Live repro on v2026.6.5-beta.2'
  • [Bug]: Regression 2026.6.1: orphan tool.call fail-closed handling silently discards the composed assistant reply ("Embedded agent failed before reply") openclaw/openclaw#91067 — beta-only: '2026.6.5-beta.2: condition unchanged'
  • [BUG] Feishu streaming card overwrites text instead of appending (#87896 regression) openclaw/openclaw#91401 — fixed on main; reporter asked when fix lands; no stable 2026.6.5 broken proof

---

IsItStable.com — score-first release stability evidence