OPENCLAW_RELEASE_TRACKER_LOG.md
OP-A and Operations and Trackers/OPENCLAW_RELEASE_TRACKER_LOG.md
# OpenClaw Release Tracker Log Append-only log of nightly OpenClaw release-alignment checks. Newest entry on top. --- ## 2026-05-24 23:56 UTC — Nightly check - Alignment doc covers through: **2026.5.18** (2026-05-19, per DOC11 R15 Amendment Proposal R3 + OPENCLAW_RELEASE_ALIGNMENT_RUNNING_SPEC_UPDATE_LIST R3, both in `Addenda Proposals & In Progress/DOC11 Openclaw Addenda & Docs/`). UNCHANGED — alignment-doc files not modified per task constraints; the R3 pair remains the operative basis. The R3 pair is now FIVE days behind the wild. - Latest stable release: **2026.5.22** (2026-05-24 01:12 UTC, signed by `steipete` Peter Steinberger, confirmed via direct GitHub release-page fetch). NEW STABLE today — second stable patch after the 5.19 promotion cut, three days after 5.20. - Gap status: **BEHIND** — 3 new stable releases since the R3-pair coverage (2026.5.19 + 2026.5.20 + 2026.5.22). Gap deepens from yesterday's BEHIND-by-2. - New since last covered: **2026.5.19** (2026-05-20, carry-forward — beta-lineage promotion, see 2026-05-21 entry), **2026.5.20** (2026-05-21, carry-forward — exec-approval skill-load contract break + Policy plugin + Codex 0.132.0 + voice realtime bootstrapContextFiles defaults + plugin slash-command autocomplete + before/after_compaction hook timeouts + doctor warnings + xAI device-code OAuth + OpenRouter params.provider routing, see 2026-05-22 entry), AND **2026.5.22** (2026-05-24 — first patch after the 5.20 cut). 2026.5.22 summary: (a) NEW Meeting Notes — source-only external meeting-notes plugin and Plugin SDK source-provider contract outside the core npm package, with auto-start capture config, manual transcript imports, read-only `openclaw meeting-notes` CLI access, and Discord voice as the first live source (NEW channel/source category — DOC11/DOC4/DOC25 ingestion-contract touchpoint). (b) NEW Plugin SDK `embeddingProviders` capability contract and registration API so embeddings become a reusable provider surface outside memory-specific adapters (DOC11 provider/model contract + ELNOR Qwen3-Embedding-0.6B MLX locked-infra interaction — does NOT migrate the locked infra, only widens the OpenClaw-side seam). (c) NEW Plugin SDK row-level session workflow helpers; `loadSessionStore` deprecated so plugins read/patch sessions without depending on the legacy whole-store shape. (d) Sub-agent bootstrap default narrowed to AGENTS.md + TOOLS.md only — persona, identity, user, memory, heartbeat, and setup files are now kept OUT of delegated workers by default (SUBAGENT V4 + DOC73/DOC1 living-memory carve-out DIRECT INTERSECTION — paired with the 5.20 voice realtime `bootstrapContextFiles` IDENTITY.md/USER.md/SOUL.md default, this is the second context-injection-surface change in three releases). (e) Gateway/perf: catalog/plugin metadata caching, lazy-load startup-idle plugin work + core gateway method handlers + embedded ACPX runtime, pre-warm provider auth-state map at startup so `/models` and every model-listing call short-circuits per-provider plugin / external-CLI discovery on the hot path (per-call cost drops ~20s → ~5ms, ~4,100×). (f) Providers/xAI+Grok: reuse xAI OAuth auth profiles for Grok `web_search`, thread active-agent auth through web search, add Grok model aliases, media providers can declare default operation timeouts. (g) Channel plugins: generic channel-message poll sender so channel plugins can expose poll delivery without depending on channel-specific SDK facades. (h) Agents/tools: REMOVES the old sender-owner tool gating path so configured tools stay visible for trusted sessions while command and channel-action auth still carry real sender identity (contract change in tool-auth gating). (i) Media understanding: stop auto-probing Gemini CLI; use Antigravity CLI only as a lower-priority image/video fallback after configured provider APIs. (j) Maintainer skills: NEW `openclaw-landable-bug-sweep` for producing five small, reviewed, CI-green OpenClaw bugfix PRs from issue/PR sweeps. (k) CLI/onboarding: classic onboarding starts when bare `openclaw` runs before an authored config exists, configured installs stay on Crestodian. (l) Models: prune retired Groq, GitHub Copilot, OpenAI, xAI, and old Claude catalog entries with doctor migration. (m) Anthropic 1M context: migrate to GA-capable Claude 4.x models without the retired `context-1m-2025-08-07` beta. (n) Codex app-server: restart and retry once when server-side compaction times out so preflight compaction stalls recover instead of failing every dispatch. (o) Release/security: ship the root npm package and OpenClaw-owned npm plugins with generated shrinkwrap, support bundled plugin runtime dependencies for suitable plugin tarballs, require review for lockfile/shrinkwrap changes so published installs use locked dependency graphs. (p) ~60 docs-touchpoint clarifications (Signal/Telegram/Termux/Bitwarden/Peekaboo Bridge/Z.AI consolidation/etc.). (q) Diagnostics: OpenTelemetry log bodies behind explicit content capture, Prometheus/observability smoke aliases. (r) Mac app native settings sidebar window chrome. WATCH-ONLY: `v2026.5.24-beta.1` (visible in releases listing as a prerelease) — do not canon unless promoted. - Most significant new items: (a) **NEW Meeting Notes plugin + Plugin SDK source-provider contract** — opens an entirely new ingestion category (Discord voice as the first live source) outside the core npm package; DOC25 must classify and DOC11 must register the new contract. (b) **Subagent bootstrap default narrowed to AGENTS.md + TOOLS.md only** — DIRECT contract change to delegated-worker context. Paired with the 5.20 voice realtime `bootstrapContextFiles` IDENTITY.md/USER.md/SOUL.md default, this is the SECOND context-injection-surface change in three releases — they should be addressed together with a single explicit DOC73/DOC1 boundary reaffirmation, not as two separate item rows. (c) **NEW Plugin SDK `embeddingProviders` capability contract** — interacts with ELNOR's LOCKED Qwen3-Embedding-0.6B MLX infra invariant. Does NOT migrate the locked infra; widens only the OpenClaw-side seam. DOC11 should make this read-only on the ELNOR side and reaffirm the locked-infra-migration-required rule. (d) NEW Plugin SDK row-level session workflow helpers + `loadSessionStore` deprecation (DOC11 session-store contract). (e) Pre-warm provider auth-state map at startup, ~4,100× speedup on `/models` and model-listing calls (DOC11 model-listing performance contract). (f) REMOVES old sender-owner tool gating path (DOC11 tool-auth contract change — DOC15 sender-scoped tool authority interaction). (g) Anthropic 1M context migration to GA Claude 4.x without the retired beta (DOC11 Anthropic provider truth). (h) Codex app-server compaction-timeout retry (DOC11 compaction owner/trigger readback + DOC24 KDA compaction-hook interaction). Carryover from 5.19 + 5.20 (exec-approval skill-load contract break, NEW Policy plugin family, per-agent `localModelLean`, Codex harness 0.132.0, voice realtime bootstrapContextFiles defaults, plugin slash-command autocomplete, before/after_compaction hook timeouts, doctor warnings, xAI device-code OAuth, OpenRouter params.provider routing) remains in scope. - Notes: WebSearch query (`OpenClaw v2026.5 release notes site:github.com/openclaw/openclaw`) returned the v2026.5.22 release page as the latest stable tag, alongside v2026.5.20, v2026.5.19, v2026.5.18, v2026.5.24-beta.1, and v2026.5.18-beta.1. Direct fetch of `https://github.com/openclaw/openclaw/releases/tag/v2026.5.22` confirmed: published `24 May 01:12 UTC`, signed by `steipete` (Peter Steinberger). Direct fetch of `v2026.5.19` and `v2026.5.22` both exceeded the per-call token limit; used Grep against the saved fetch buffers to extract the `### Changes` and release-date sections. v2026.5.20 fetch succeeded inline as a smaller payload. The 3-stable-cut gap (5.19 + 5.20 + 5.22) plus the second context-injection-surface change in three releases now make the patch-vs-roll-forward decision a roll-forward: standing recommendation ESCALATED from "R3 Patch 1 addendum" to a small **DOC11 R15 R4 roll-forward** folding ALL THREE 5.19 + 5.20 + 5.22 releases in one pass before the R14 → R15 redline is drafted. The two highest-priority items are (1) the exec-approval skill-load contract break (5.20) and (2) the coordinated subagent + voice-realtime context-injection-surface boundary statement (5.20 + 5.22). The embedding-providers capability is a NEW reaffirmation point for the locked Qwen3-Embedding-0.6B MLX infra invariant. Affected spec touchpoints (updated): DOC11 (PRIMARY — exec-approval skill-load contract break, Policy plugin family, per-agent localModelLean, Codex harness 0.132.0, NEW Meeting Notes plugin/SDK source-provider contract, NEW embeddingProviders capability, NEW row-level session workflow helpers, REMOVED sender-owner tool gating, /models pre-warm performance contract, plugin slash-command autocomplete, before/after_compaction hook timeouts, doctor warnings, Anthropic 1M context migration), DOC4 (Plugin SDK contract surface widening — source-provider, embeddingProviders, row-level session helpers, generic channel-message poll sender), DOC15 (TTS pre-send ordering carryover from 5.19; sender-scoped tool authority interaction with REMOVED sender-owner gating), DOC24 (compaction hook timeout owner/trigger interaction with 5.22 Codex compaction-timeout retry; voice realtime bootstrap context files as identity-context surface intersecting DOC72/DOC1; Meeting Notes as new KDA-relevant ingestion category), DOC25 (Meeting Notes ingestion category — NEW classification needed; sandbox_exec/sandbox_process carryover from 5.19; Discord voice-channel-follow ingestion gate carryover from 5.20), DOC72/DOC1/DOC73 (CONSOLIDATED context-injection-surface boundary reaffirmation needed — covers BOTH the 5.20 voice realtime bootstrapContextFiles IDENTITY.md/USER.md/SOUL.md default AND the 5.22 subagent bootstrap narrowing to AGENTS.md+TOOLS.md only; the embedding-providers capability is also an indirect DOC72 vector-store-contract touchpoint via Qwen3 locked infra), EC Core (Policy plugin as high-authority plugin-runtime register from 5.20; exec-approval contract break in skill-load path from 5.20; admin HTTP RPC web QR login carryover from 5.19; npm shrinkwrap + bundled plugin runtime dependency policy from 5.22 as installer/supply-chain truth), SUBAGENT V4 (DIRECT IMPACT — subagent bootstrap default narrowed to AGENTS.md + TOOLS.md is a contract change to delegated-worker context; Codex harness 0.132.0 + clean-bounded-refactor carryover from 5.19). Dashboard updates today: `latestReleaseChecked` → `2026.5.22`; `latestReleaseCheckedDate` → `2026-05-24`; `releasesSinceLastCovered` extended with `2026.5.22` entry (5.19 + 5.20 entries retained); `gapStatus` remains `BEHIND` (now BEHIND-by-3); `lastChecked` → `2026-05-24`; `recommendation` rewritten to flag BEHIND-by-3, the R3 → R4 roll-forward path anchored on the exec-approval skill-load contract break PLUS the consolidated context-injection-surface boundary statement, the new Meeting Notes + embeddingProviders + row-level session SDK contract surface, and the affected spec touchpoints; `alignmentDocLastCovered` / `alignmentDocDate` / `alignmentDocFiles` preserved (unchanged). Alignment-doc files themselves not modified per task constraints. WATCH: v2026.5.24-beta.1 (visible in releases listing as prerelease — do not canon unless promoted before redline drafting); any further 2026.5.x stable patches in the next 24-48h (the 5.20 → 5.22 cadence was 3 days; a 5.23 / 5.24 stable cut within the week is plausible). --- ## 2026-05-22 10:42 UTC — Nightly check - Alignment doc covers through: **2026.5.18** (2026-05-19, per DOC11 R15 Amendment Proposal R3 + OPENCLAW_RELEASE_ALIGNMENT_RUNNING_SPEC_UPDATE_LIST R3, both in `Addenda Proposals & In Progress/DOC11 Openclaw Addenda & Docs/`). UNCHANGED from yesterday — alignment-doc files not modified per task constraints; the R3 pair remains the operative basis. - Latest stable release: **2026.5.20** (2026-05-21 20:44 UTC, confirmed via direct GitHub release-page fetch — page is tagged `[Latest]`, `[Latest]`, 117 commits behind main as of release time, signed by `steipete` Peter Steinberger). NEW STABLE today — one stable patch behind the 5.19 promotion cut tracked in yesterday's entry. - Gap status: **BEHIND** — 2 new stable releases since the R3-pair coverage (2026.5.19 + 2026.5.20). Gap deepens from yesterday's BEHIND-by-1. - New since last covered: **2026.5.19** (2026-05-20, carry-forward from yesterday's entry — beta-lineage promotion: clean-bounded-refactor agent defaults, @openclaw/proxyline 0.3.3, Pi 0.75.1, Node 22.19 minimum, NEW OPENCLAW_IMAGE_APT_PACKAGES build-arg, Gateway/ACPX restart-trace cost attribution, overlapped startup logging + plugin-service startup with channel sidecars preserving /readyz gating, NEW admin-http-rpc QR-login flow, Codex app-server sandbox_exec/sandbox_process for non-Docker SSH sandbox backends, stale session diagnostics + Codex OAuth fallback recovery, Messages/TTS directives applied before message-tool sends reach core/gateway/plugin delivery) AND **2026.5.20** (2026-05-21 — first patch after the 5.19 promotion). 2026.5.20 summary: (a) Exec approvals — REMOVES the legacy `cat SKILL.md && printf ... && <skill-wrapper>` allowlist compatibility path; skill files must now be loaded via the read tool and only the real skill executable is auto-allowed (contract break vs. R3-era assumptions about skill setup exec-approval). (b) NEW bundled Policy plugin for policy-backed channel conformance checks, doctor lint findings, and opt-in workspace repair. (c) Codex harness bumped to @openai/codex 0.132.0 with refreshed app-server model-list docs. (d) NEW per-agent `agents.list[].experimental.localModelLean` (replaces global-only flag). (e) Providers/xAI — device-code OAuth login for remote/headless setups (no localhost browser callback). (f) Providers/OpenRouter — honors provider-level `params.provider` routing policy with model/agent params overriding defaults. (g) Discord — voice sessions can follow configured users into voice channels (allowed-channel checks, multi-user handoff, DAVE recovery preservation); realtime sessions include bounded IDENTITY.md/USER.md/SOUL.md profile context in instructions by default, with `voice.realtime.bootstrapContextFiles: []` opt-out. (h) NEW 30s default timeout on `before_compaction`/`after_compaction` plugin hooks so a hung handler no longer blocks compaction. (i) Plugin slash commands now show in TUI autocomplete. (j) Doctor — warns on plaintext secret-bearing config fields (model provider API keys, sensitive provider headers) and when sandbox tool policy hides configured MCP server tools. (k) Restored fail-closed `tryReadSecretFileSync` `rejectSymlink: true` contract (Telegram/LINE/Zalo/IRC/Nextcloud Talk credential loaders refuse symlinked secret files). (l) Cron — main-session scheduled work runs on a cron-owned wake lane preserving reply delivery context, structured embedded-run denial metadata for isolated scheduled tasks, legacy top-level array `jobs.json` preserved across upgrade. (m) Browser — honors configured image sanitization limit for screenshots/labeled snapshots. (n) ~40+ additional fixes spanning Codex, agents/subagents, Gateway, WhatsApp, Mac app, iOS, installer, channels, CLI. WATCH-ONLY: `v2026.5.19-beta.2` remains on the beta dist-tag — likely fully subsumed by 5.20 stable but any beta-only content needs verification before next R. - Most significant new items: (a) **Exec-approval skill-load contract break** — the legacy `cat SKILL.md && printf ... && <skill-wrapper>` allowlist compatibility path is REMOVED; skill files must be loaded via the read tool and only the real skill executable is auto-allowed (highest-priority single item — DOC25/DOC73 callouts that assumed the legacy contract for skill setup auth need an explicit migration path). (b) NEW bundled Policy plugin family (channel conformance + doctor lint + workspace repair) — new plugin-runtime register that DOC11 R15 must enumerate. (c) Discord/voice realtime `bootstrapContextFiles` IDENTITY.md/USER.md/SOUL.md default context — a NEW context-injection surface that intersects DOC72/DOC1/DOC73 living-memory vs. static-facts boundary (must NOT erode the carve-outs). (d) Per-agent `experimental.localModelLean` (DOC11 model/provider control-surface readback). (e) Codex harness 0.132.0 bump + app-server model-list refresh (DOC11 Codex truth). (f) `before_compaction`/`after_compaction` plugin hook 30s default timeout (DOC11 compaction owner/trigger/context-window readback + DOC24 KDA delivery-directive interaction). (g) Plugin slash commands in TUI autocomplete (DOC11 command catalog truth). (h) Doctor plaintext-secret + sandbox-MCP-hidden warnings (diagnostics surface). (i) Restored `tryReadSecretFileSync` symlink fail-closed (infra/secrets contract). 5.19's prior items (admin-http-rpc QR login, Codex app-server sandbox_exec/sandbox_process, Gateway/ACPX restart-trace cost attribution, OPENCLAW_IMAGE_APT_PACKAGES build-arg, Node 22.19 minimum, Messages/TTS pre-send directive ordering) remain in scope. - Notes: WebSearch query (`OpenClaw v2026.5 release notes site:github.com/openclaw/openclaw`) returned the v2026.5.20 release page as the `[Latest]` tag, with v2026.5.19 and v2026.5.19-beta.2 also visible alongside v2026.5.18 and v2026.5.18-beta.1. Direct fetch of `https://github.com/openclaw/openclaw/releases/tag/v2026.5.20` confirmed: published `21 May 20:44`, 117 commits behind main as of release time, signed by `steipete` (Peter Steinberger). Direct fetch of `v2026.5.19` failed due to result size exceeding token limit; relied on WebSearch snippet for that release's summary (already captured in yesterday's entry). The 2-stable-cut gap (5.19 + 5.20) makes a deeper roll-forward more attractive than a one-release patch. Standing recommendation reinforced: rather than blocking the DOC11 R14 → R15 redline, draft a **DOC11 R15 R3 → R4 roll-forward (or single "R3 Patch 1" addendum)** folding BOTH 5.19 and 5.20 in one targeted pass — the exec-approval skill-load contract break is the highest-priority single item and should anchor the patch. Affected spec touchpoints: DOC11 (primary — exec-approval skill-load contract break, NEW Policy plugin family, per-agent localModelLean, Codex harness 0.132.0, plugin slash-command autocomplete, before/after_compaction hook timeouts, doctor warnings, xAI device-code OAuth + OpenRouter params.provider routing, all 5.19 carryover items), DOC15 (TTS pre-send ordering carryover from 5.19), DOC24 (KDA delivery-directive interaction with compaction-hook timeouts; voice-realtime bootstrapContextFiles as identity-context surface; proactive-send voice-note carryover from 5.19), DOC25 (sandbox_exec/sandbox_process ingestion path carryover from 5.19; Discord voice-channel-follow ingestion gate), DOC72/DOC1/DOC73 (realtime voice `bootstrapContextFiles` IDENTITY.md/USER.md/SOUL.md is a NEW context-injection surface that must NOT erode living-memory vs. static-facts boundary — direct intersection with PBE carve-outs and DOC73/DOC1 boundary discipline), EC Core (Policy plugin as high-authority plugin-runtime register; exec-approval skill-load contract break; admin-http-rpc QR-login flow carryover from 5.19), SUBAGENT V4 (Codex harness 0.132.0 + 5.19's clean-bounded-refactor + lean-internals + plugin-SDK-API deprecation-path agent defaults). Dashboard updates today: `latestReleaseChecked` → `2026.5.20`; `latestReleaseCheckedDate` → `2026-05-22`; `releasesSinceLastCovered` extended with `2026.5.20` entry (5.19 entry retained); `gapStatus` remains `BEHIND` (now BEHIND-by-2); `lastChecked` → `2026-05-22`; `recommendation` rewritten to flag BEHIND-by-2, the R3 → R4 roll-forward path (or R3 Patch 1 addendum) anchored on the exec-approval skill-load contract break, and the affected spec touchpoints; `alignmentDocLastCovered` / `alignmentDocDate` / `alignmentDocFiles` preserved (unchanged). Alignment-doc files themselves not modified per task constraints. WATCH: any further 2026.5.x stable patches in the next 24-48h (5.20 ships 117 commits behind main as of release time, so a 5.21 cut is plausible); v2026.5.19-beta.2 if it carries beta-only items not in 5.20 stable. --- ## 2026-05-21 10:45 UTC — Nightly check - Alignment doc covers through: **2026.5.18** (2026-05-19, per DOC11 R15 Amendment Proposal R3 + OPENCLAW_RELEASE_ALIGNMENT_RUNNING_SPEC_UPDATE_LIST R3, both in `Addenda Proposals & In Progress/DOC11 Openclaw Addenda & Docs/`). UNCHANGED from yesterday — alignment-doc files not modified per task constraints; the R3 pair remains the operative basis. - Latest stable release: **2026.5.19** (estimated 2026-05-20 publish date — promotion of the v2026.5.19-beta.1/.2 lineage to stable; per WebSearch the v2026.5.19 tag now appears alongside v2026.5.18 on the openclaw/openclaw GitHub releases page, with v2026.5.19-beta.2 also visible as a separate prerelease tag). NEW STABLE today — ends the 40-hour 5.18 stable-channel quiet period and the brief two-consecutive-CURRENT entries in this log (5/19 + 5/20). - Gap status: **BEHIND** — 1 new stable release (2026.5.19). First reopened gap since the R3 pair closed the 22-day cadence lag on 2026-05-19. - New since last covered: **2026.5.19** (~2026-05-20). Summary: (1) Agents — fixes default to clean bounded refactors + lean internals + explicit plugin SDK/API deprecation paths. (2) Dependencies — `@openclaw/proxyline` 0.3.3, Pi packages 0.75.1, raised minimum supported Node.js 22 line to 22.19 (carry-forward from 5.19-beta.1). (3) Docker/Podman — NEW `OPENCLAW_IMAGE_APT_PACKAGES` runtime-neutral image build arg for extra apt packages while keeping `OPENCLAW_DOCKER_APT_PACKAGES` as a legacy fallback. (4) Gateway/ACPX — attribute startup probe, config, runtime, and resource-count costs in restart traces without changing readiness behavior; overlap startup logging + plugin-service startup with channel sidecars to reduce restart ready latency while preserving `/readyz` sidecar gating. (5) Plugins/admin-http-rpc — NEW capability allowing trusted admin HTTP RPC clients to start and wait for web QR login flows. (6) Codex app-server — expose OpenClaw's sandbox-routed shell as `sandbox_exec`/`sandbox_process` for non-Docker sandbox backends so SSH sandbox agents keep a correctly routed shell path without shadowing Codex native shell. (7) Release stability — recover stale session diagnostics + Codex OAuth fallback state so stuck runs and reused refresh tokens clear without blocking follow-up work. (8) Messages/TTS — apply TTS directives before message-tool sends reach core/gateway/plugin delivery so opt-in message-tool rooms and proactive sends attach voice notes instead of leaking raw tags. WATCH-ONLY beta channel: `v2026.5.19-beta.2` also appears on the beta dist-tag — likely fully subsumed by the 5.19 stable cut, but any beta-only content needs verification before next R. - Most significant new items: (a) Gateway/ACPX restart-trace cost attribution lines (architecturally a quiet diagnostics readback contract that DOC11 should canonize alongside the existing diagnostics/restart trace surfaces); (b) admin HTTP RPC web QR login flow (a high-authority pairing/auth surface — EC Core secrets/pairing governance + DOC11 plugin admin surface intersection); (c) Codex app-server `sandbox_exec`/`sandbox_process` for non-Docker sandbox backends (DOC11 Codex provider truth + DOC25 ingestion contract for SSH sandbox agents); (d) Messages/TTS directive ordering before message-tool delivery (DOC15 message-tool send ordering + DOC24 proactive-send voice-note directive ordering); (e) `OPENCLAW_IMAGE_APT_PACKAGES` runtime-neutral image build arg (DOC11 container-image build-arg catalog); (f) Node 22.19 minimum + `@openclaw/proxyline` 0.3.3 + Pi 0.75.1 (DOC11 runtime-environment minimums readback). The R3 pair already covered the surrounding 2026.4.27 → 2026.5.18 wave; 5.19 is best handled as a small targeted roll-forward (R3 Patch 1 / R4 minor) rather than reopening the consolidated wave. - Notes: WebSearch queries (`OpenClaw v2026.5 release notes site:github.com/openclaw/openclaw`, `openclaw 2026.5.19 release notes site:github.com`, `openclaw 2026.5.20 OR 2026.5.21 release notes github`, `"openclaw" "2026.5.20" release github`) all corroborate: latest stable tag is now `v2026.5.19` (promotion of beta.1/.2 lineage; appears alongside `v2026.5.18` on the GitHub releases page; one source noted npm latest may still lag at `v2026.5.18` for a brief sync window). No `v2026.5.20` of any form observed. No `v2026.5.21` of any form. Two-day post-5.18 stable pause that held through 5/19 + 5/20 ended with this targeted 5.19 cut, consistent with the typical "large stable release → 1-7 day pause → patch or watch-pause-extension" post-stable patch window. Standing recommendation: rather than blocking the DOC11 R14 → R15 redline on a new R4, draft a small **DOC11 R15 R3 Patch 1** (or R4 minor) folding 2026.5.19 in one targeted pass — keeping R3-as-base + roll-forward delta rather than re-opening the consolidated wave; companion patch to the running list. Affected spec touchpoints: DOC11 (primary — gateway/ACPX restart-trace lines, admin-http-rpc QR-login flow, Codex app-server sandbox_exec/sandbox_process, OPENCLAW_IMAGE_APT_PACKAGES build-arg, Node 22.19 minimum, dependency floors), DOC15 (TTS-before-message-tool-send ordering — opt-in message-tool rooms voice-note attachment, separate from DOC24's proactive-send directive ordering), DOC24 (proactive-send voice-note directive ordering — touches KDA delivery-directive contract), DOC25 (sandbox_exec/sandbox_process ingestion-contract path for SSH sandbox agents), EC Core (admin HTTP RPC web QR login flow as high-authority pairing/auth surface), SUBAGENT V4 (clean-bounded-refactor + lean-internals + plugin-SDK-API deprecation-path agent defaults — quiet contract reinforcement). Dashboard updates today: `latestReleaseChecked` → '2026.5.19'; `latestReleaseCheckedDate` → '2026-05-21'; `releasesSinceLastCovered` populated with the 2026.5.19 entry; `gapStatus` → 'BEHIND' (reopened from yesterday's CURRENT); `lastChecked` → '2026-05-21'; `recommendation` rewritten to flag BEHIND-by-1, the R3 Patch 1 / R4 minor proposal path, and the affected spec touchpoints; `alignmentDocLastCovered` / `alignmentDocDate` / `alignmentDocFiles` preserved (unchanged). Alignment-doc files themselves not modified per task constraints. --- ## 2026-05-20 10:41 UTC — Nightly check - Alignment doc covers through: **2026.5.18** (2026-05-19, per DOC11 R15 Amendment Proposal R3 + OPENCLAW_RELEASE_ALIGNMENT_RUNNING_SPEC_UPDATE_LIST R3, both in `Addenda Proposals & In Progress/DOC11 Openclaw Addenda & Docs/`). UNCHANGED from yesterday — the R3 pair remains the operative alignment basis; no new alignment-doc revision today. - Latest stable release: **2026.5.18** (2026-05-18 18:54 UTC, per WebSearch — Codex app-server fix to keep native code-mode available without forcing code-mode-only so OpenClaw dynamic tool turns complete through the app-server tool bridge; recovery of stale session diagnostics and Codex OAuth fallback state so stuck runs and reused refresh tokens clear without blocking follow-up work). UNCHANGED from yesterday — now T+2 days old. - Gap status: **CURRENT** — alignment remains CAUGHT UP. Second consecutive CURRENT entry in this log (yesterday was the first). The twenty-two-day single-release roll-forward lag that had been tracked since 2026-04-26 remains formally closed. - New since last covered: **none stable**. No new stable OpenClaw release has been published between 2026-05-18 18:54 UTC and 2026-05-20 10:41 UTC (~40 hours of stable-channel quiet). Beta channel: `v2026.5.19-beta.1` (published 2026-05-18) remains the latest pre-release; no new beta tag observed since yesterday. The 5.16-beta lineage (beta.1 → beta.3 → beta.4 → beta.5 tracked through 5/17) remains overtaken and consolidated into the 5.18 stable cut + 5.19-beta.1 forward, with no 5.13/5.14/5.15/5.16/5.17 stable cuts materialized. - Most significant new items: none today — R3 still IS the refresh. The R3 pair continues to absorb the entire 2026.4.27 → 2026.5.18 stable wave (externalized plugin lifecycle, plugin-owned Gateway method registries, typed tool-plugin authoring `defineToolPlugin` / `plugins build` / `validate` / `init`, active-run interaction `/steer` / `/btw` / `/side` / follow-up / collect / interrupt queue modes, branch metadata readback without false branch-chat-UI claims, dual browser architecture, browser dialog/action/URL allowlist + modal-dialog handling + evaluate timeout + CDP port/profile state + browser artifact gating, provider-native + plugin-backed web search with SSRF/private-network/proxy controls and DOC72 ingest gates, Codex app-server/code-mode/native-tool/sandbox-network controls and legacy `codex-cli/*` migration truth, native voice / Talk / TTS / STT / realtime + Android Gateway relay voice, compaction owner/trigger/context-window/source-ref readback + pre-compaction ELNOR capture protection, diagnostics/export/restart/config reloadability/provider quota/session.tool progress/readiness traces, paired-node file-transfer controls, sender-scoped tool authority, model/provider auth + quota + source plan + fallback + catalog repair + per-agent bootstrap profile controls, streaming/progress/final-delivery normalization). The Codex OAuth route saga remains closed in safe three-act form (5.5 → 5.6 → 5.7). - Notes: WebSearch queries (`OpenClaw v2026.5 release notes site:github.com/openclaw/openclaw`, `github.com openclaw openclaw releases v2026.5.20 OR v2026.5.19 OR v2026.5.18 stable May 2026`, `"openclaw" "2026.5.20" OR "2026.5.21" release github`) all corroborate: latest stable tag remains `v2026.5.18` (2026-05-18T18:54Z, prerelease=false); latest pre-release remains `v2026.5.19-beta.1` (2026-05-18, prerelease=true). No `v2026.5.20` / `v2026.5.21` of any form observed. Two-day stable pause since 5.18 is well within the normal post-release patch window (typical: large stable release → 1-7 day pause → patch or watch-pause-extension). Standing recommendation: proceed with drafting the DOC11 R14 → R15 redline from R3 (the proposal remains coding-handoff ready); keep R3 watch-only rows out of canon until `v2026.5.19-beta.1` either promotes to stable or a new stable release lands. WATCH: if a new stable release lands in the next 2-3 days the gap will reopen at BEHIND-by-1 — likely a tighter roll-forward rather than another consolidated wave, given the R3 pair just absorbed the prior backlog. Dashboard updates today: `latestReleaseCheckedDate` → '2026-05-20'; `lastChecked` → '2026-05-20'; `recommendation` rewritten to reflect today's no-new-stable T+2 finding, second consecutive CURRENT entry, 40-hour stable-channel quiet since 5.18, and the watch trigger for if/when a new stable lands; `alignmentDocLastCovered` / `alignmentDocDate` / `alignmentDocFiles` / `latestReleaseChecked` / `releasesSinceLastCovered` / `gapStatus` all preserved (unchanged). Alignment-doc files themselves not modified per task constraints. --- ## 2026-05-19 10:46 UTC — Nightly check - Alignment doc covers through: **2026.5.18** (2026-05-19, per BRAND-NEW DOC11 R15 Amendment Proposal R3 + OPENCLAW_RELEASE_ALIGNMENT_RUNNING_SPEC_UPDATE_LIST R3 both shipped today). The R3 pair supersedes R2 in full and absorbs the entire 2026.4.27 → 2026.5.18 stable wave in one consolidated pass. - Latest stable release: **2026.5.18** (2026-05-18, per WebSearch — Codex app-server fix to keep native code-mode available without forcing code-mode-only so OpenClaw dynamic tool turns complete through the app-server tool bridge, and recovery of stale session diagnostics and Codex OAuth fallback state so stuck runs and reused refresh tokens clear without blocking follow-up work). No new stable release has been published between 2026-05-18 and 2026-05-19 10:46 UTC. - Gap status: **CURRENT** — alignment now CAUGHT UP. The twenty-two-day single-release roll-forward lag tracked since 2026-04-26 is now formally closed. This is the first CURRENT entry in this log; every prior entry has been BEHIND. - New since last covered: none stable. Beta channel: `v2026.5.19-beta.1` published 2026-05-18 (clean bounded refactor defaults, lean internals, explicit plugin SDK/API deprecation paths; @openclaw/proxyline 0.3.3, Pi 0.75.1, Node 22.19 minimum). Tracked as `watch_only` per R3 §0.1 release-status rule — will not be canonized in the DOC11 R15 redline unless promoted to stable before redline drafting. Yesterday's 5.16-beta lineage (beta.1 → beta.3 → beta.4 → beta.5) appears to have been overtaken/consolidated into the 5.18 stable cut + 5.19-beta.1 forward; no 5.13/5.14/5.15/5.16/5.17 stable cuts ever materialized. - Most significant new items: the R3 alignment pair lands an extensive consolidated amendment covering externalized plugin lifecycle, plugin-owned Gateway method registries, typed tool-plugin authoring (`defineToolPlugin` / `plugins build` / `validate` / `init`), active-run interaction (`/steer`, `/btw`, `/side`, follow-up, collect, interrupt queue modes), branch metadata readback (without falsely claiming a complete branch-chat UI), dual browser architecture (Q Browser internal intake vs OpenClaw/Chrome/CDP/extension automation), browser dialog/action/URL allowlist + modal-dialog handling + evaluate timeout + CDP port/profile state + browser artifact gating, provider-native + plugin-backed web search with SSRF/private-network/proxy controls and DOC72 ingest gates, Codex app-server/code-mode/native-tool/sandbox-network controls and legacy `codex-cli/*` migration truth, native voice / Talk / TTS / STT / realtime + Android Gateway relay voice, compaction owner/trigger/context-window/source-ref readback + pre-compaction ELNOR capture protection, diagnostics/export/restart/config reloadability/provider quota/session.tool progress/readiness traces, paired-node file-transfer controls, sender-scoped tool authority, model/provider auth + quota + source plan + fallback + catalog repair + per-agent bootstrap profile controls, and streaming/progress/final-delivery normalization. The Codex OAuth route saga remains closed in safe three-act form (5.5 → 5.6 → 5.7); 2026.5.12 leaner-installs + `acp.fallbacks` + pairing-before-proxy + Claude-CLI bounded-transcript retry + in-process agent dispatcher all absorbed; 2026.5.18 Codex app-server tool-bridge fix + session diagnostics recovery added as the most recent stable item. - Notes: WebSearch queries (`OpenClaw v2026.5 release notes site:github.com/openclaw/openclaw`, `openclaw 2026.5.19 release github`) corroborate latest stable = `v2026.5.18` (2026-05-18, prerelease=false) and beta = `v2026.5.19-beta.1` (2026-05-18, prerelease=true). The shipping of R3 today closes the long-running 22-day roll-forward lag in a single coordinated proposal; the architect should now treat R3 as coding-handoff-ready and proceed with drafting the actual DOC11 R14 → R15 redline. Spec touchpoints absorbed: DOC11 (primary owner), DOC4 (plugin architecture), DOC3 (skills/tool-plugin authoring), DOC10 (orchestration ledger / room residue split), DOC12 (channel projection), DOC15 (Active Memory + admin-scope), DOC20/DOC22 (UI surfaces — auto-scroll, branch UX disclaimers), DOC24 (ingestion + KDA + Codex/browser allowlist gates), DOC25 (ingestion contracts), EC Core (secrets/pairing + high-authority file-transfer + paired-node controls), SUBAGENT V4 (sub-agent precedence + `agents.defaults.subagents.model`). Dashboard updates today: `alignmentDocLastCovered` → `2026.5.18`; `alignmentDocDate` → `2026-05-19`; `alignmentDocFiles` rewritten to the new R3 paths under `Addenda Proposals & In Progress/DOC11 Openclaw Addenda & Docs/` (R2 files are now in that subfolder's `Archived/`); `latestReleaseChecked` → `2026.5.18`; `latestReleaseCheckedDate` → `2026-05-19`; `releasesSinceLastCovered` → `[]`; `gapStatus` → `CURRENT`; `lastChecked` → `2026-05-19`; `recommendation` rewritten to reflect CURRENT status + R3-as-the-refresh + watch_only handling for 2026.5.19-beta.1. --- ## 2026-05-18 10:44 — Nightly check - Alignment doc covers through: 2026.4.26 (2026-04-29, per DOC11 R15 R2 amendment proposal — newer than running list R2 at 2026-04-13 / 2026.4.12) - Latest stable release: 2026.5.12 (2026-05-14, published 18:28 UTC) — UNCHANGED from yesterday's check, now T+4 days old. The 5/14 stable cut that ended the 7-day pause continues to hold without a follow-up patch (no 5.12.1 / 5.13 / 5.14 / 5.15 / 5.16 / 5.17 stable cut). - Gap status: BEHIND (10 new releases — 2026.4.27, 2026.4.29, 2026.5.2, 2026.5.3, 2026.5.3-1, 2026.5.4, 2026.5.5, 2026.5.6, 2026.5.7, 2026.5.12; 2026.4.28 / 2026.4.30 / 2026.5.0 / 2026.5.1 / 2026.5.8 / 2026.5.9 / 2026.5.10 / 2026.5.11 / 2026.5.13 / 2026.5.14 / 2026.5.15 / 2026.5.16 / 2026.5.17 not published as stable — 5.9 / 5.10 / 5.14 / 5.16 exist only as betas, others not at all). UNCHANGED in composition from yesterday's 5/17 entry (10 entries). - New since last covered: none stable. No new stable OpenClaw release has been published since 2026.5.12 (2026-05-14 18:28 UTC). WATCH-ONLY beta-channel activity since yesterday: the previously-tracked `v2026.5.16-beta.3` beta dist-tag has rolled forward to a NEW `v2026.5.16-beta.4` prerelease (per Releasebot snippet — stability, auth, cron, gateway updates, xAI Grok OAuth login for SuperGrok subscribers carried forward, smarter cron blocking for queued runs, better session and config handling, localized setup flows, many fixes for media/chat/provider reliability) and then to a NEW `v2026.5.16-beta.5` prerelease (latest, last published 2026-05-17 per npm versions snippet; details not yet inspected line-by-line). Beta dist-tag only — none promote into the stable alignment basis until cut on the latest dist-tag. The 5.14-beta lineage (tracked through 5.14-beta.2 on 5/15) remains ABANDONED — no v2026.5.13 / v2026.5.14 stable promotion. The 5.16-beta lineage cadence (beta.1 → beta.3 → beta.4 → beta.5 across roughly 2026-05-15 → 2026-05-17) is now tracking like the 5.12-beta lineage that ran beta.1 → beta.4 → beta.5 → beta.6 → stable over ~5 days, so a stable promotion (most likely as v2026.5.15 / v2026.5.16 / v2026.5.17) within the next 2-3 days is the leading hypothesis. The version-number sequence now skips v2026.5.13, v2026.5.14, AND v2026.5.15 entirely at the stable tier; the beta sequence runs 5.12-beta.6 → 5.14-beta.1 → 5.14-beta.2 → 5.16-beta.1 → 5.16-beta.3 → 5.16-beta.4 → 5.16-beta.5 (5.13 and 5.15 absent at every tier; 5.14 present at beta only). GitHub issue #82069 documenting v2026.5.14-beta.1 Codex OAuth refresh 401 token_expired regression remains tracked — its resolution status in 5.16-beta.5 is not yet confirmed via direct fetch. - Most significant new items: none stable. The 10-release backlog narrative is unchanged from yesterday. The 2026.5.12 architectural anchors (A through H from prior entries) remain the new-this-cycle items: leaner-installs plugin-externalization extension (Bedrock/Bedrock Mantle/Slack/OpenShell sandbox/Anthropic Vertex out of core); NEW `acp.fallbacks` ACP runtime-backend fallback capability; explicit browser device pairing + Control UI pairing before proxy-scoped access; Claude CLI fresh-session bounded transcript retry; in-process agent dispatcher for same-process subagent handoffs; Control UI/WebChat persisted auto-scroll mode selector; Telegram polling isolation + durable local spool + HTML formatting preservation; `agents.defaults.subagents.model` application before target primary models during `sessions_spawn`. Codex OAuth route saga remains closed in safe three-act form (5.5 unsafe rewrite → 5.6 revert → 5.7 safe preserve/recover); DOC11 R15 R3 may treat the conditional Codex-route preservation rule as stable precedent. WATCH items from beta channel (pre-stable, not yet alignment-basis material): xAI Grok OAuth-login-for-SuperGrok carried through the 5.16-beta.3/4/5 lineage intersects DOC11 provider auth catalog (a new OAuth source for xAI/Grok parallel to OpenAI/Codex and Anthropic OAuth flows) + DOC10/DOC11 always-on collaboration-mode contract if it stabilizes; smarter cron-runs blocking with `openclaw cron run --wait` + timeout/poll-interval controls + `cron.runs --run-id` filtering intersects DOC11 cron CLI truth (extending the 2026.5.7 cron status field quiet contract change); localized setup flows and Linux user systemd bus environment recovery from VNC/stripped-desktop-shell terminals from earlier 5.16-beta entries remain DOC11 setup/onboarding readback items; the previously-tracked v2026.5.14-beta.2 Channels/SDK normalized command-turn-facts → channel-turn-construction + command-turn helpers for plugin inbound contexts may or may not be carried forward into the 5.16-beta lineage (needs verification once 5.16-beta full release notes inspected); the open v2026.5.14-beta.1 Codex OAuth refresh 401 token_expired issue #82069 status post-5.16-beta.5 is unknown. No new architectural surfaces today, no new schema obligations, no new Codex/route changes, no new plugin lifecycle changes in stable. - Notes: WebSearch queries (`OpenClaw v2026 release notes site:github.com/openclaw/openclaw`, `OpenClaw stable release 2026.5 changelog github.com/openclaw/openclaw releases`, `openclaw 2026.5.13 OR 2026.5.14 OR 2026.5.15 OR 2026.5.16 OR 2026.5.17 stable release github tag`, `openclaw "2026.5.16-beta.4" OR "2026.5.16-beta.5" release notes`, plus prior queries for stable 5.12 confirmation) all corroborate: latest stable tag remains `v2026.5.12` (2026-05-14T18:28Z, prerelease=false). New observations today vs yesterday: `v2026.5.16-beta.4` and `v2026.5.16-beta.5` prereleases now visible on the beta dist-tag (5.16-beta.5 last published 2026-05-17 per npm versions snippet); the 5.14-beta lineage remains abandoned. No `v2026.5.13` of any form; no `v2026.5.14` stable; no `v2026.5.15` of any form; no `v2026.5.16` stable; no `v2026.5.17` of any form. Four-day stable pause since 5.12 remains within the post-architectural-release patch/feature window typical for OpenClaw (large stable release → 1-7 day pause → patch or watch-pause-extension) and consistent with the beta-channel pattern of skipping 5.13/5.15 and aggregating into 5.16. Single-release roll-forwards have now lagged behind release cadence for **twenty-two consecutive days**; the four-day stable pause since 5.12 does not narrow the gap, only halts its growth. Standing strong recommendation to actually start drafting DOC11 R15 R3 consolidated amendment continues to compound — R3 should fold all ten backlog releases in one consolidated pass, with the Codex-route saga folded as a single closed three-act item rather than three separate rows; the seven 2026.5.7 quiet contract changes (Active Memory admin-scope, Channels CLI restructuring, Cron CLI status field, native command owner enforcement, auto-reply skill-tool authorization, Codex pre-guardian PermissionRequest removal + allow-always memory, Discord voice permission auditing) as DOC11 readback-truth obligations; 2026.5.12 leaner-installs + `acp.fallbacks` + pairing-before-proxy + Claude-CLI-bounded-transcript-retry + in-process-agent-dispatcher as the new architectural anchors; companion next-R of the running list (R2 → R3) carries the same logic. WATCH: 5.16-beta.5 promotion to stable (likely as `v2026.5.15`, `v2026.5.16`, or `v2026.5.17` within the next 2-3 days based on the 5.12 beta lineage cadence of beta.1 → beta.4 → beta.5 → beta.6 → stable over ~5 days) would push backlog to 11 and make the R3 pass time-sensitive; the 5.16 stable cut is likely substantively larger than a roll-forward patch since it folds both the 5.14-beta and 5.16-beta lineages; xAI Grok OAuth-login-for-SuperGrok is the watch-row most likely to require new DOC11 provider-auth catalog work once stabilized. Worth tracking whether issue #82069 (Codex OAuth refresh 401 token_expired post-#80738) is closed by the 5.16-beta.5 lineage before its stable promotion. Dashboard updates today: `lastChecked` → '2026-05-18'; `recommendation` rewritten to reflect today's no-new-stable fourth-consecutive-quiet-day finding, T+4-day-old 5.12 stable, 22-day cadence lag, 5.16-beta lineage rolled forward to beta.4 then beta.5, revised watch trigger (5.16-beta.5 promotion to 5.15/5.16/5.17 stable within next 2-3 days); `releasesSinceLastCovered` / `latestReleaseChecked` / `latestReleaseCheckedDate` / `gapStatus` all preserved (unchanged); alignment-doc fields preserved per task constraints (alignment docs themselves are not modified). --- ## 2026-05-17 10:44 — Nightly check - Alignment doc covers through: 2026.4.26 (2026-04-29, per DOC11 R15 R2 amendment proposal — newer than running list R2 at 2026-04-13 / 2026.4.12) - Latest stable release: 2026.5.12 (2026-05-14, published 18:28 UTC) — UNCHANGED from yesterday's check, now T+3 days old. The 5/14 stable cut that ended the 7-day pause continues to hold without a follow-up patch (no 5.12.1 / 5.13 / 5.14 / 5.15 stable cut). - Gap status: BEHIND (10 new releases — 2026.4.27, 2026.4.29, 2026.5.2, 2026.5.3, 2026.5.3-1, 2026.5.4, 2026.5.5, 2026.5.6, 2026.5.7, 2026.5.12; 2026.4.28 / 2026.4.30 / 2026.5.0 / 2026.5.1 / 2026.5.8 / 2026.5.9 / 2026.5.10 / 2026.5.11 / 2026.5.13 / 2026.5.14 / 2026.5.15 not published as stable — 5.9 / 5.10 / 5.14 / 5.16 exist only as betas, others not at all). UNCHANGED in composition from yesterday's 5/16 entry (10 entries). - New since last covered: none stable. No new stable OpenClaw release has been published since 2026.5.12 (2026-05-14 18:28 UTC). WATCH-ONLY beta-channel activity since yesterday: yesterday's tracked `v2026.5.14-beta.2` (2026-05-15 11:11 UTC) appears ABANDONED — the beta dist-tag has rolled forward to a NEW `v2026.5.16-beta.1` and then a NEW `v2026.5.16-beta.3` prerelease tag (latter published roughly today per WebSearch snippet — adds xAI Grok OAuth login for SuperGrok subscribers, emphasizes tool-contract / collaboration-mode / runtime-reliability hardening, removes "small failure modes that make always-on assistants annoying or unsafe in real teams"; details not yet inspected line-by-line). The yesterday-tracked 24-72h watch for 5.14-beta.2 → stable v2026.5.13 / v2026.5.14 did NOT materialize; the beta lineage instead jumped 5.14-beta.2 → 5.16-beta.1 → 5.16-beta.3, suggesting the next stable cut will be substantively larger (folding two beta lineages) rather than a 5.13 roll-forward patch. Beta dist-tag only — does not promote into the stable alignment basis until cut on the latest dist-tag. The version-number sequence now skips `v2026.5.13`, `v2026.5.14`, AND `v2026.5.15` entirely at the stable tier; the beta sequence runs 5.12-beta.6 → 5.14-beta.1 → 5.14-beta.2 → 5.16-beta.1 → 5.16-beta.3 (5.13 and 5.15 absent at every tier; 5.14 present at beta only). The GitHub issue #82069 documenting `v2026.5.14-beta.1` Codex OAuth refresh 401 token_expired regression remains tracked — its resolution status in 5.16-beta.3 is not yet confirmed via direct fetch. - Most significant new items: none stable. The 10-release backlog narrative is unchanged from yesterday. The 2026.5.12 architectural anchors (A through H from the 5/15 entry) remain the new-this-cycle items: leaner-installs plugin-externalization extension (Bedrock/Bedrock Mantle/Slack/OpenShell sandbox/Anthropic Vertex out of core); NEW `acp.fallbacks` ACP runtime-backend fallback capability; explicit browser device pairing + Control UI pairing before proxy-scoped access; Claude CLI fresh-session bounded transcript retry; in-process agent dispatcher for same-process subagent handoffs; Control UI/WebChat persisted auto-scroll mode selector; Telegram polling isolation + durable local spool + HTML formatting preservation; `agents.defaults.subagents.model` application before target primary models during `sessions_spawn`. Codex OAuth route saga remains closed in safe three-act form (5.5 unsafe rewrite → 5.6 revert → 5.7 safe preserve/recover); DOC11 R15 R3 may treat the conditional Codex-route preservation rule as stable precedent. WATCH items from beta channel (pre-stable, not yet alignment-basis material): the new `v2026.5.16-beta.3` xAI Grok OAuth-login-for-SuperGrok and "always-on assistant failure mode" hardening intersect DOC11 provider auth catalog (specifically: a new OAuth source for xAI/Grok parallel to the existing OpenAI/Codex and Anthropic OAuth flows) + DOC10/DOC11 always-on collaboration-mode contract if it stabilizes; the yesterday-tracked `v2026.5.14-beta.2` Channels/SDK normalized command-turn-facts → channel-turn-construction + command-turn helpers for plugin inbound contexts may or may not be carried forward into the 5.16-beta lineage (needs verification once 5.16-beta full release notes inspected); the open `v2026.5.14-beta.1` Codex OAuth refresh 401 token_expired issue #82069 status post-5.16-beta.3 is unknown. No new architectural surfaces today, no new schema obligations, no new Codex/route changes, no new plugin lifecycle changes in stable. - Notes: WebSearch queries (`OpenClaw v2026 release notes site:github.com/openclaw/openclaw May 2026`, `OpenClaw 2026.5 latest stable release github openclaw/openclaw`, `OpenClaw stable release tag "2026.5.13" OR "2026.5.14" OR "2026.5.15" OR "2026.5.16" OR "2026.5.17" github`, `"openclaw 2026.5.16-beta" OR "openclaw 2026.5.16" release date May 16 17`, plus prior queries for stable 5.12 confirmation) all corroborate: latest stable tag remains `v2026.5.12` (2026-05-14T18:28Z, prerelease=false). New observations today vs yesterday: `v2026.5.16-beta.1` and `v2026.5.16-beta.3` prereleases now visible on beta dist-tag; the 5.14-beta lineage (tracked through 5.14-beta.2 yesterday) has been overtaken without stable promotion; the yesterday-noted 24-72h promotion-to-5.13/5.14-stable hypothesis did not materialize. No `v2026.5.13` of any form; no `v2026.5.14` stable; no `v2026.5.15` of any form. Three-day stable pause since 5.12 is still within the normal post-architectural-release patch/feature window (typical pattern: large stable release → 1-7 day pause → patch or watch-pause-extension) and now consistent with the beta-channel pattern of skipping 5.13/5.15 and aggregating into 5.16. Single-release roll-forwards have now lagged behind release cadence for **twenty-one consecutive days**; the three-day stable pause does not narrow the gap, only halts its growth. Standing strong recommendation to actually start drafting DOC11 R15 R3 consolidated amendment continues to compound — R3 should fold all ten backlog releases in one consolidated pass, with the Codex-route saga folded as a single closed three-act item rather than three separate rows; the seven 2026.5.7 quiet contract changes (Active Memory admin-scope, Channels CLI restructuring, Cron CLI status field, native command owner enforcement, auto-reply skill-tool authorization, Codex pre-guardian PermissionRequest removal + allow-always memory, Discord voice permission auditing) as DOC11 readback-truth obligations; 2026.5.12 leaner-installs + `acp.fallbacks` + pairing-before-proxy + Claude-CLI-bounded-transcript-retry + in-process-agent-dispatcher as the new architectural anchors; companion next-R of the running list (R2 → R3) carries the same logic. WATCH: if `v2026.5.16-beta.3` is promoted to stable (most likely as `v2026.5.15`, `v2026.5.16`, or `v2026.5.17` within the next 2-5 days based on the 5.12 beta lineage cadence which ran beta.1 → beta.4 → beta.5 → beta.6 → stable over ~5 days), backlog jumps to 11 and the R3 pass becomes more time-sensitive — and the 5.16 stable cut is likely to be substantively larger than a roll-forward patch given it folds both 5.14-beta and 5.16-beta lineages; also worth tracking whether issue #82069 (Codex OAuth refresh 401 token_expired post-#80738) is closed by the 5.16-beta lineage before its stable promotion. Dashboard updates today: `lastChecked` → '2026-05-17'; `recommendation` rewritten to reflect today's no-new-stable third-consecutive-quiet-day finding, T+3-day-old 5.12 stable, 21-day cadence lag, abandoned 5.14-beta lineage and beta-channel roll-forward to 5.16-beta.1 then 5.16-beta.3, revised watch trigger (5.16-beta.3 promotion to 5.15/5.16/5.17 stable rather than 5.14-beta.2 promotion to 5.13/5.14 stable); `releasesSinceLastCovered` / `latestReleaseChecked` / `latestReleaseCheckedDate` / `gapStatus` all preserved (unchanged); alignment-doc fields preserved per task constraints (alignment docs themselves are not modified). --- ## 2026-05-16 10:43 — Nightly check - Alignment doc covers through: 2026.4.26 (2026-04-29, per DOC11 R15 R2 amendment proposal — newer than running list R2 at 2026-04-13 / 2026.4.12) - Latest stable release: 2026.5.12 (2026-05-14, published 18:28 UTC) — UNCHANGED from yesterday's check, now T+2 days old. The 5/14 stable cut that ended the 7-day pause has held without an immediate follow-up patch (no same-day 5.12.1 or 5.13 cut). - Gap status: BEHIND (10 new releases — 2026.4.27, 2026.4.29, 2026.5.2, 2026.5.3, 2026.5.3-1, 2026.5.4, 2026.5.5, 2026.5.6, 2026.5.7, 2026.5.12; 2026.4.28 / 2026.4.30 / 2026.5.0 / 2026.5.1 / 2026.5.8 / 2026.5.9 / 2026.5.10 / 2026.5.11 / 2026.5.13 not published as stable — 5.9 / 5.10 / 5.14 exist only as betas, others not at all). UNCHANGED in composition from yesterday's 5/15 entry (10 entries). - New since last covered: none stable. No new stable OpenClaw release has been published since 2026.5.12 (2026-05-14 18:28 UTC). WATCH-ONLY beta-channel activity since yesterday: **NEW `v2026.5.14-beta.2` prerelease tag dated 2026-05-15 11:11 UTC** (per WebSearch snippet — additions to Channels/SDK for normalized command turn facts to channel turn construction and command-turn helpers for plugin inbound contexts, plus dependency refreshes, Canvas updates, UI/i18n touches, Telegram support changes, and assorted agent functionality improvements; details not yet inspected line-by-line) on the beta dist-tag, following the previously-tracked `v2026.5.14-beta.1` (2026-05-14/15 — Codex OAuth refresh 401 token_expired investigation thread #82069 reports issue post-#80738 against beta.1, then beta.2 ships). Beta dist-tag only — does not promote into the stable alignment basis until cut on the latest dist-tag. The version-number sequence skips `v2026.5.13` entirely so far at both stable and beta level (5.12 → 5.14-beta.1 → 5.14-beta.2, no 5.13 of any form observed). - Most significant new items: none stable. The 10-release backlog narrative is unchanged from yesterday. The 2026.5.12 architectural anchors (A through H from yesterday's entry) remain the new-this-cycle items: leaner-installs plugin-externalization extension (Bedrock/Bedrock Mantle/Slack/OpenShell sandbox/Anthropic Vertex out of core); NEW `acp.fallbacks` ACP runtime-backend fallback capability; explicit browser device pairing + Control UI pairing before proxy-scoped access; Claude CLI fresh-session bounded transcript retry; in-process agent dispatcher for same-process subagent handoffs; Control UI/WebChat persisted auto-scroll mode selector; Telegram polling isolation + durable local spool + HTML formatting preservation; `agents.defaults.subagents.model` application before target primary models during `sessions_spawn`. Codex OAuth route saga remains closed in safe three-act form (5.5 unsafe rewrite → 5.6 revert → 5.7 safe preserve/recover); DOC11 R15 R3 may treat the conditional Codex-route preservation rule as stable precedent. WATCH items from beta channel (pre-stable, not yet alignment-basis material): the new `v2026.5.14-beta.2` Channels/SDK normalized command-turn-facts → channel-turn-construction + command-turn helpers for plugin inbound contexts intersect DOC11 channels truth + DOC4 plugin contract (specifically: how channel inbound paths surface command-turn facts to plugin handlers — a quiet contract for any spec that pins plugin inbound surface) if it stabilizes; the open `v2026.5.14-beta.1` Codex OAuth refresh 401 token_expired issue #82069 indicates the post-#80738 Codex OAuth subscriber refresh path may not be fully closed yet — relevant to DOC11 Codex provider truth + EC Core secrets/pairing governance if the eventual stable cut canonizes a fix. No new architectural surfaces today, no new schema obligations, no new Codex/route changes, no new plugin lifecycle changes in stable. - Notes: WebSearch queries (`OpenClaw v2026.5 release notes site:github.com/openclaw/openclaw May 2026`, `OpenClaw v2026.5.13 OR v2026.5.14 OR v2026.5.15 release site:github.com/openclaw/openclaw`, `openclaw 2026.5.13 stable release github`) plus partial fetch of github.com/openclaw/openclaw/releases (top page output exceeded inline render budget; per saved-file scan the visible page index contained only entries back through v2026.5.4 and v2026.5.5 was labeled `[Latest]` in the rendered HTML — likely a stale cached snapshot of the releases page, since direct WebSearch links resolve to v2026.5.12 release page and v2026.5.14-beta.2 release page) confirm latest stable tag remains `v2026.5.12` (2026-05-14T18:28Z, prerelease=false). New observation today: `v2026.5.14-beta.2` (2026-05-15 11:11 UTC, prerelease) and the GitHub issue #82069 documenting `v2026.5.14-beta.1` Codex OAuth refresh regression report. No `v2026.5.13` of any form observed; no `v2026.5.14` stable; no `v2026.5.15` of any form. Two-day stable pause since 5.12 is well within the normal post-architectural-release patch window (typical pattern: large stable release → 1-3 day pause → patch or watch-pause-extension). Single-release roll-forwards have now lagged behind release cadence for **twenty consecutive days**; the 5/14 stable cut resumed gap growth (8→9→10) and today's no-change holds it at 10. Standing strong recommendation to actually start drafting DOC11 R15 R3 consolidated amendment continues to compound — R3 should fold all ten backlog releases in one consolidated pass, with the Codex-route saga folded as a single closed three-act item rather than three separate rows; companion next-R of the running list (R2 → R3) carries the same logic. WATCH: if `v2026.5.14-beta.2` is promoted to stable (most likely as `v2026.5.13` or `v2026.5.14` within the next 24-72h based on the 5.12 beta lineage cadence of beta.1 → beta.4 → beta.5 → beta.6 → stable taking ~5 days), backlog jumps to 11 and the R3 pass becomes more time-sensitive; also worth tracking whether issue #82069 (Codex OAuth refresh 401 token_expired post-#80738) reopens the Codex-route saga in stable form before R3 lands. Dashboard updates today: `lastChecked` → '2026-05-16'; `recommendation` rewritten to reflect today's no-new-stable finding, T+2-day-old 5.12 stable, 20-day cadence lag, new `v2026.5.14-beta.2` beta watch row plus the open #82069 Codex OAuth refresh issue, and the watch trigger for if/when 5.14-beta.2 promotes to stable; `releasesSinceLastCovered` / `latestReleaseChecked` / `latestReleaseCheckedDate` / `gapStatus` all preserved (unchanged); alignment-doc fields preserved per task constraints (alignment docs themselves are not modified). --- ## 2026-05-15 10:43 — Nightly check - Alignment doc covers through: 2026.4.26 (2026-04-29, per DOC11 R15 R2 amendment proposal — newer than running list R2 at 2026-04-13 / 2026.4.12) - Latest stable release: **2026.5.12 (2026-05-14, published 18:28 UTC by github-actions, tag signed by steipete)** — NEW STABLE today, ending the 7-day stable pause that ran 2026-05-08 → 2026-05-14. The 1649-commits-to-main count visible at fetch time confirms healthy stable-tag cut. - Gap status: BEHIND (**10 new releases** — 2026.4.27, 2026.4.29, 2026.5.2, 2026.5.3, 2026.5.3-1, 2026.5.4, 2026.5.5, 2026.5.6, 2026.5.7, AND **2026.5.12**; 2026.4.28 / 2026.4.30 / 2026.5.0 / 2026.5.1 / 2026.5.8 / 2026.5.9 / 2026.5.10 / 2026.5.11 were not published as stable — 5.9 / 5.10 existed only as betas, 5.11 not at all, version sequence jumped 5.7 → 5.12 stable). Backlog composition grew by one for the first time since the 5/8 entry (5/8 was the last day with growth from 8→9). - New since last covered: **2026.5.12** (2026-05-14). Highlights confirmed from direct fetch of github.com/openclaw/openclaw/releases/tag/v2026.5.12: (1) **Leaner installs** — WhatsApp, Slack, Amazon Bedrock, Anthropic Vertex, and related provider/plugin dependency cones moved out of the core runtime so installs only pull what you use (extends 2026.5.2 plugin-externalization: Bedrock + Bedrock Mantle externalized to separate provider packages; Slack + OpenShell sandbox + Anthropic Vertex externalized). (2) **Telegram resilience** — isolated polling worker with durable local spool (#81132); preserved HTML/Markdown formatting in streamed and lazy cron-announced replies (#81742); skip unmentioned group media before download when requireMention is active (#81181); tool-progress drafts kept separate from final assistant answer. (3) **Codex/OpenAI smoother** — auth-profile-backed media tools stay available when OpenAI auth stored outside env vars; MCP server projection; context-engine thread rotation; better app-server/runtime fallback. (4) **Plugin install/update hardening** — pnpm 11 support; peer-dependency preservation; safer runtime scans; source/git install fixes; installer honors --version for git installs and installs from checked-in lockfile. (5) **Gateway/browser/Slack/node-pairing/sandbox/transcript security-provenance pass** — requires explicit browser device pairing and Control UI pairing before proxy-scoped access. (6) **UI/reply delivery** — Control UI/WebChat persisted auto-scroll mode selector with three modes (current near-bottom, always-follow streaming, off + manual New-messages button) (#7648, #81287); rich-only reply paths, session history, streaming auto-scroll touched. (7) **NEW ACP capability** — `acp.fallbacks` lets ACP turns try configured backup runtime backends when primary backend is unavailable before any output is emitted (#69542). (8) **Agents/subagents** — apply `agents.defaults.subagents.model` before target agent primary models during `sessions_spawn` so model-scoped runtimes such as claude-cli stay attached to default child runs (#81395). (9) **Agents** — deliver same-process subagent completion handoffs through in-process agent dispatcher instead of opening a Gateway RPC loopback; add permissive item schemas to array tool parameters before provider submission. (10) **Claude CLI fresh-session retries** — bounded OpenClaw transcript history after session rotation, preventing conversation amnesia (#80905). (11) **Channels/Weixin** — bump external Weixin catalog entry to `@tencent-weixin/openclaw-weixin@2.4.3` with matching package integrity (#81730). (12) **CLI/plugins** — keep bare plugin and parent-command help on the lightweight path, avoiding plugin registry discovery before rendering help. Beta lineage that converged: 5.12-beta.1 → 5.12-beta.4 → 5.12-beta.5 → 5.12-beta.6 → stable 5.12; many of the previously-tracked 5.9/5.10/5.12-beta watch items (iMessage media-only send hygiene from 5.12-beta.6; Codex private-runtime-SDK fix from 5.12-beta.4; /think default + /fast default from 5.9-beta.1; openclaw.json 0600 chmod from 5.9-beta.1; etc.) are NOT individually called out in the 5.12 stable highlights — they may have landed quietly or been deferred; needs direct verification against full 5.12 changelog rather than the abridged highlights view. - Most significant new items: **2026.5.12** is the most architecturally consequential stable release since 2026.5.3. Top architectural anchors for the next R3 amendment pass: (A) Leaner-installs plugin-externalization extension (Bedrock/Bedrock Mantle/Slack/OpenShell sandbox/Anthropic Vertex out of core) — DOC4 plugin architecture + DOC11 provider/plugin ownership truth; (B) NEW `acp.fallbacks` ACP runtime-backend fallback capability — DOC11 ACP core ownership (directly affects RLD-12-01 durable channel/topic bindings cluster and RLD-11-REF-04 ACP core setup/runtime); (C) Explicit browser device pairing + Control UI pairing before proxy-scoped access — DOC11 + EC Core secrets/pairing governance security-posture contract; (D) Claude CLI fresh-session bounded transcript retry preventing conversation amnesia — DOC11 model/runtime truth + sub-agent contract; (E) In-process agent dispatcher for same-process subagent handoffs — DOC10 orchestration ledger + DOC11 sub-agent runtime truth; (F) Control UI/WebChat persisted auto-scroll mode selector — DOC20/DOC22 UI surface; (G) Telegram polling isolation + durable local spool + HTML formatting preservation — DOC12 channel projection; (H) `agents.defaults.subagents.model` application before target primary models during `sessions_spawn` — SUBAGENT V4 + DOC11 sub-agent precedence. Codex OAuth route saga remains closed in safe three-act form (5.5 unsafe rewrite → 5.6 revert → 5.7 safe preserve/recover); DOC11 R15 R3 may treat the conditional Codex-route preservation rule as stable precedent. - Notes: Direct fetch of github.com/openclaw/openclaw/releases/tag/v2026.5.12 succeeded today (100,596 bytes; output exceeded inline render budget but parseable via saved file); confirmed publish date `released this 14 May 18:28` on the latest dist-tag with `[Latest]` label and verified-signature commit by steipete. WebSearch queries (`OpenClaw v2026 release notes site:github.com/openclaw/openclaw May 2026`, `OpenClaw 2026.5.12 release notes stable github`, `OpenClaw latest stable release github 2026.5 May 14 15`, `site:github.com openclaw releases tag v2026.5.12`, `OpenClaw 2026.5.12 release date stable highlights leaner installs Telegram WhatsApp`) all corroborate: latest stable is now `v2026.5.12` (2026-05-14, prerelease=false). No `v2026.5.13` or `v2026.5.14` of any form yet. No new betas appeared on the beta dist-tag between yesterday's check and now (or none observed in today's WebSearch result snippets). RECOMMENDATION ESCALATED: 7-day stable pause has ended with a substantively architectural release, so the standing strong recommendation to draft DOC11 R15 R3 (and the companion running-list R3) now compounds with one more release to fold. R3 should fold all ten backlog releases in one consolidated pass — Codex-route saga as a single closed three-act item; the seven 2026.5.7 quiet contract changes as DOC11 readback-truth obligations; 2026.5.12 leaner-installs + `acp.fallbacks` + pairing-before-proxy + Claude-CLI-bounded-transcript-retry + in-process-agent-dispatcher as the new architectural anchors. Single-release roll-forwards have now lagged behind release cadence for **eighteen consecutive days**; the new stable resumes gap growth after the 7-day stable pause. Dashboard updates today: `latestReleaseChecked` → '2026.5.12'; `latestReleaseCheckedDate` → '2026-05-14'; `releasesSinceLastCovered` array appended with 2026.5.12 entry; `lastChecked` → '2026-05-15'; `recommendation` rewritten to reflect the new stable, escalated R3-drafting case, ten-release backlog, eighteen-day cadence lag, and 2026.5.12 architectural anchors; `gapStatus` preserved as 'BEHIND' (composition changed 9 → 10); alignment-doc fields preserved per task constraints (alignment docs themselves are not modified). --- ## 2026-05-14 10:42 — Nightly check - Alignment doc covers through: 2026.4.26 (2026-04-29, per DOC11 R15 R2 amendment proposal — newer than running list R2 at 2026-04-13 / 2026.4.12) - Latest stable release: 2026.5.7 (2026-05-07, published 20:57 UTC) — UNCHANGED for the seventh consecutive day. The 2026.4.27 → 2026.5.7 daily-cadence streak has now paused for 7 days. - Gap status: BEHIND (9 new releases — 2026.4.27, 2026.4.29, 2026.5.2, 2026.5.3, 2026.5.3-1, 2026.5.4, 2026.5.5, 2026.5.6, 2026.5.7; 2026.4.28 / 2026.4.30 / 2026.5.0 / 2026.5.1 / 2026.5.8 were not published as stable). UNCHANGED in composition from 2026-05-08 / 2026-05-09 / 2026-05-10 / 2026-05-12 / 2026-05-13. - New since last covered: none today. No new stable OpenClaw release has been published since 2026.5.7 (2026-05-07 20:57 UTC). This is the **seventh consecutive day** without a new stable release. WATCH-ONLY beta-channel activity continued today: the previously-tracked `v2026.5.9-beta.1` (2026-05-09 — `/think default` + `/fast default` chat-command additions that clear session overrides and inherit configured/provider defaults; POSIX `openclaw.json` 0600 chmod restoration after suspicious-read backup recovery), `v2026.5.10-beta.6` (2026-05-11), and `v2026.5.12-beta.4` (2026-05-13 05:17 UTC — Codex private-runtime-SDK helper exposure fix for `@openclaw/codex` MODULE_NOT_FOUND errors) are now joined by a NEW `v2026.5.12-beta.6` prerelease tag dated 2026-05-13 (per WebSearch snippet — iMessage native image send: stop emitting visible `<media:image>` placeholder text for media-only sends while preserving the internal echo key that prevents self-echo duplicate replies). Beta dist-tag only — none of the prereleases promote into the stable alignment basis until cut on the latest dist-tag. The version-number sequence continues to skip `v2026.5.8` AND `v2026.5.11` entirely; the beta sequence runs 5.9-beta.1 → 5.10-beta.6 → 5.12-beta.4 → 5.12-beta.6. The 5.12-beta.6 contents are only an iMessage visible-text hygiene fix — no new architectural surfaces, beta cadence not yet alignment-basis material. - Most significant new items: none stable. The 9-release backlog narrative is unchanged from the 5/13 entry. Codex OAuth route saga remains closed in safe three-act form (2026.5.5 unsafe rewrite → 2026.5.6 revert → 2026.5.7 safe preserve/recover); DOC11 R15 R3 may treat the conditional Codex-route preservation rule as stable precedent for capability-bound default-runtime resolution. 2026.5.3's bundled file-transfer plugin + gateway invalid-config guards remain the most schema-consequential items; 2026.5.2's plugin externalization (ACPX, diagnostics-otel, npm/ClawHub publishing prep) remains the most architecturally consequential plugin-packaging shift; 2026.4.27's Codex Computer Use + DeepInfra entrance + LINE inbound media path remain the watch → include-now promotion candidates; the seven 2026.5.7 quiet contract changes (Active Memory admin-scope #78863, Channels CLI restructuring #78456, Cron CLI status field #78701, native command owner enforcement #78864, auto-reply skill-tool authorization #78517, Codex pre-guardian PermissionRequest removal + allow-always session memory, Discord voice permission auditing) remain DOC11 readback-truth obligations. WATCH items from beta channel (pre-stable, not yet alignment-basis material): the new `v2026.5.12-beta.6` iMessage media-only-send hygiene fix touches DOC11/DOC10 channel projection (visible-reply enforcement vs internal echo key separation — quiet contract for any spec that pins iMessage visible-reply truth) if it stabilizes; `v2026.5.12-beta.4` Codex private-runtime-SDK fix touches DOC11 Codex provider + plugin-runtime SDK-exposure truth if it stabilizes; `/think default` + `/fast default` session-override clearing semantics from 5.9-beta.1 intersect DOC11 native-command catalog + session-override floor truth; `openclaw.json` 0600 chmod restoration after suspicious-read backup recovery is a credential-handling hardening item touching EC Core secrets governance + DOC11 backup/recovery truth; `v2026.5.10-beta.6` contents not yet enumerated in detail. No new architectural surfaces today, no new schema obligations, no new Codex/route changes, no new plugin lifecycle changes in stable. - Notes: WebSearch queries (`OpenClaw v2026.5 release notes site:github.com/openclaw/openclaw May 2026`, `OpenClaw v2026.5.8 OR v2026.5.9 OR v2026.5.10 OR v2026.5.11 OR v2026.5.12 stable release github.com/openclaw`, `"openclaw" "v2026.5.12-beta" OR "v2026.5.13" OR "v2026.5.14" release github`) plus partial fetch of github.com/openclaw/openclaw/releases (output exceeded inline render budget; release listing visible only via download) confirm latest stable tag is still `v2026.5.7` (2026-05-07T20:57:43Z, fixes-only, prerelease=false). New observation today: `v2026.5.12-beta.6` (2026-05-13, prerelease — iMessage visible `<media:image>` placeholder text removal while preserving self-echo key) on top of the already-known `v2026.5.12-beta.4` (2026-05-13, prerelease), `v2026.5.10-beta.6` (2026-05-11, prerelease), and `v2026.5.9-beta.1` (2026-05-09, prerelease). No `v2026.5.8` of any form, no `v2026.5.11` of any form, no `v2026.5.13`/`v2026.5.14` of any form. Seven-day stable pause remains unusual against the prior 2026.4.27 → 2026.5.7 daily cadence and further supports the standing strong recommendation to **actually start drafting** the DOC11 R15 R3 consolidated amendment during this pause rather than wait for further releases that would all need rolling together. Single-release roll-forwards have now lagged behind release cadence for seventeen consecutive days; the seven-day pause does not narrow the gap, only halts its growth. R3 should fold 2026.4.27 (promote watch → include-now), 2026.4.29, 2026.5.2, 2026.5.3, 2026.5.3-1, 2026.5.4, 2026.5.5, 2026.5.6, AND 2026.5.7 in one consolidated pass — with the Codex-route saga folded as a single closed three-act item rather than three separate rows. Companion next-R of the running list (R2 → R3) carries the same logic. Dashboard updates today: `lastChecked` → 2026-05-14; `recommendation` rewritten to reflect today's no-change stable finding, seventh-consecutive-quiet-day status, seventeen-day cadence lag, and new `v2026.5.12-beta.6` beta watch row alongside `v2026.5.12-beta.4` / `v2026.5.10-beta.6` / `v2026.5.9-beta.1`; `releasesSinceLastCovered` / `latestReleaseChecked` / `latestReleaseCheckedDate` / `gapStatus` all preserved (unchanged); alignment-doc fields preserved per task constraints (alignment docs themselves are not modified). --- ## 2026-05-13 10:43 — Nightly check - Alignment doc covers through: 2026.4.26 (2026-04-29, per DOC11 R15 R2 amendment proposal — newer than running list R2 at 2026-04-13 / 2026.4.12) - Latest stable release: 2026.5.7 (2026-05-07, published 20:57 UTC) — UNCHANGED for the sixth consecutive day. The 2026.4.27 → 2026.5.7 daily-cadence streak has now paused for 6 days. - Gap status: BEHIND (9 new releases — 2026.4.27, 2026.4.29, 2026.5.2, 2026.5.3, 2026.5.3-1, 2026.5.4, 2026.5.5, 2026.5.6, 2026.5.7; 2026.4.28 / 2026.4.30 / 2026.5.0 / 2026.5.1 / 2026.5.8 were not published as stable). UNCHANGED in composition from 2026-05-08 / 2026-05-09 / 2026-05-10 / 2026-05-12. - New since last covered: none today. No new stable OpenClaw release has been published since 2026.5.7 (2026-05-07 20:57 UTC). This is the **sixth consecutive day** without a new stable release. WATCH-ONLY beta-channel activity has continued and added a new prerelease today: the previously-tracked `v2026.5.9-beta.1` (2026-05-09 — `/think default` + `/fast default` chat-command additions that clear session overrides and inherit configured/provider defaults; POSIX `openclaw.json` 0600 chmod restoration after suspicious-read backup recovery) and `v2026.5.10-beta.6` (2026-05-11) are now joined by a NEW `v2026.5.12-beta.4` prerelease tag dated 2026-05-13 05:17 UTC (per WebSearch snippet — Codex runtime fix allowing the official installed `@openclaw/codex` package to use its private task-runtime SDK helper, addressing `MODULE_NOT_FOUND` errors during migrated OpenAI/Codex beta runs). Beta dist-tag only — none of the prereleases promote into the stable alignment basis until cut on the latest dist-tag. The version-number sequence is now skipping `v2026.5.8` AND `v2026.5.11` entirely (no stable or beta tag for either); the beta sequence runs 5.9-beta.1 → 5.10-beta.6 → 5.12-beta.4. The 5.12-beta.4 contents are not yet inspected in detail beyond the Codex runtime SDK fix — beta cadence not yet alignment-basis material. - Most significant new items: none stable. The 9-release backlog narrative is unchanged from the 5/12 entry. Codex OAuth route saga remains closed in safe three-act form (2026.5.5 unsafe rewrite → 2026.5.6 revert → 2026.5.7 safe preserve/recover); DOC11 R15 R3 may treat the conditional Codex-route preservation rule as stable precedent for capability-bound default-runtime resolution. 2026.5.3's bundled file-transfer plugin + gateway invalid-config guards remain the most schema-consequential items; 2026.5.2's plugin externalization (ACPX, diagnostics-otel, npm/ClawHub publishing prep) remains the most architecturally consequential plugin-packaging shift; 2026.4.27's Codex Computer Use + DeepInfra entrance + LINE inbound media path remain the watch → include-now promotion candidates; the seven 2026.5.7 quiet contract changes (Active Memory admin-scope #78863, Channels CLI restructuring #78456, Cron CLI status field #78701, native command owner enforcement #78864, auto-reply skill-tool authorization #78517, Codex pre-guardian PermissionRequest removal + allow-always session memory, Discord voice permission auditing) remain DOC11 readback-truth obligations. WATCH items from beta channel (pre-stable, not yet alignment-basis material): the new `v2026.5.12-beta.4` Codex private-runtime-SDK fix touches DOC11 Codex provider + plugin-runtime SDK-exposure truth if it stabilizes (specifically: the rule that an officially-installed `@openclaw/codex` package can access its own private task-runtime SDK helper — a plugin-runtime-API surface that DOC11 R15 should explicitly recognize); `/think default` + `/fast default` session-override clearing semantics from 5.9-beta.1 intersect DOC11 native-command catalog + session-override floor truth; `openclaw.json` 0600 chmod restoration after suspicious-read backup recovery is a credential-handling hardening item touching EC Core secrets governance + DOC11 backup/recovery truth; `v2026.5.10-beta.6` contents not yet enumerated in detail. No new architectural surfaces today, no new schema obligations, no new Codex/route changes, no new plugin lifecycle changes in stable. - Notes: WebSearch queries (`OpenClaw v2026.5 release notes site:github.com/openclaw/openclaw May 2026`, `"openclaw 2026.5.8" OR "openclaw 2026.5.10" OR "openclaw 2026.5.11" OR "openclaw 2026.5.12" stable release`, `openclaw "2026.5.8" stable release github tag`, `github.com/openclaw/openclaw releases latest May 13 2026`, `"openclaw 2026.5.7" OR "openclaw 2026.5.8" stable release github tag prerelease=false`) plus partial fetch of github.com/openclaw/openclaw/releases (top page returned releases back through 2026.5.4 — no newer stable tag than 2026.5.7 visible in the page index) confirm latest stable tag is still `v2026.5.7` (2026-05-07T20:57:43Z, fixes-only, prerelease=false). New observation today: `v2026.5.12-beta.4` (2026-05-13 05:17 UTC, prerelease — Codex runtime private-SDK helper exposure fix for `@openclaw/codex` MODULE_NOT_FOUND errors during migrated OpenAI/Codex beta runs) on top of the already-known `v2026.5.10-beta.6` (2026-05-11, prerelease) and `v2026.5.9-beta.1` (2026-05-09, prerelease). No `v2026.5.8` of any form, no `v2026.5.11` of any form. Six-day stable pause remains unusual against the prior 2026.4.27 → 2026.5.7 daily cadence and further supports the standing strong recommendation to **actually start drafting** the DOC11 R15 R3 consolidated amendment during this pause rather than wait for further releases that would all need rolling together. Single-release roll-forwards have now lagged behind release cadence for sixteen consecutive days; the six-day pause does not narrow the gap, only halts its growth. R3 should fold 2026.4.27 (promote watch → include-now), 2026.4.29, 2026.5.2, 2026.5.3, 2026.5.3-1, 2026.5.4, 2026.5.5, 2026.5.6, AND 2026.5.7 in one consolidated pass — with the Codex-route saga folded as a single closed three-act item rather than three separate rows. Companion next-R of the running list (R2 → R3) carries the same logic. Dashboard updates today: `lastChecked` → 2026-05-13; `recommendation` rewritten to reflect today's no-change stable finding, sixth-consecutive-quiet-day status, sixteen-day cadence lag, new `v2026.5.12-beta.4` beta watch row alongside `v2026.5.10-beta.6` and `v2026.5.9-beta.1`, and continued call to begin drafting R3 during the quiet window; `releasesSinceLastCovered` / `latestReleaseChecked` / `latestReleaseCheckedDate` / `gapStatus` all preserved (unchanged); alignment-doc fields preserved per task constraints (alignment docs themselves are not modified). --- ## 2026-05-12 15:08 — Nightly check - Alignment doc covers through: 2026.4.26 (2026-04-29, per DOC11 R15 R2 amendment proposal — newer than running list R2 at 2026-04-13 / 2026.4.12) - Latest stable release: 2026.5.7 (2026-05-07, published 20:57 UTC) — UNCHANGED for the fifth consecutive day. The 2026-05-11 nightly entry is absent from this log (tracker appears not to have run yesterday); previous log entry is 2026-05-10. The 2026.4.27 → 2026.5.7 daily-cadence streak has now been paused for 5 days. - Gap status: BEHIND (9 new releases — 2026.4.27, 2026.4.29, 2026.5.2, 2026.5.3, 2026.5.3-1, 2026.5.4, 2026.5.5, 2026.5.6, 2026.5.7; 2026.4.28 / 2026.4.30 / 2026.5.0 / 2026.5.1 / 2026.5.8 were not published as stable). UNCHANGED in composition from 2026-05-08 / 2026-05-09 / 2026-05-10. - New since last covered: none today. No new stable OpenClaw release has been published since 2026.5.7 (2026-05-07 20:57 UTC). This is the **fifth consecutive day** without a new stable release. WATCH-ONLY beta-channel activity continues: the previously-tracked `v2026.5.9-beta.1` (2026-05-09 — `/think default` + `/fast default` chat-command additions that clear session overrides and inherit configured/provider defaults; POSIX `openclaw.json` 0600 chmod restoration after suspicious-read backup recovery) is now joined by a NEW `v2026.5.10-beta.6` prerelease tag dated 2026-05-11 (per `openclaw - npm` versions index and search-result snippet "the latest version is v2026.5.10-beta.6, last published May 11, 2026"). Beta dist-tag only — neither prerelease promotes into the stable alignment basis until cut on the latest dist-tag. The version-number sequence appears to be skipping `v2026.5.8` entirely the way it skipped `v2026.4.28` / `v2026.4.30` / `v2026.5.0` / `v2026.5.1` (no `v2026.5.8` tag exists in any form, stable or beta). The 5.10-beta.6 contents are not yet inspected in detail — beta cadence not yet alignment-basis material. - Most significant new items: none stable. The 9-release backlog narrative is unchanged from the 5/10 entry. Codex OAuth route saga remains closed in safe three-act form (2026.5.5 unsafe rewrite → 2026.5.6 revert → 2026.5.7 safe preserve/recover); DOC11 R15 R3 may treat the conditional Codex-route preservation rule as stable precedent for capability-bound default-runtime resolution. 2026.5.3's bundled file-transfer plugin + gateway invalid-config guards remain the most schema-consequential items; 2026.5.2's plugin externalization (ACPX, diagnostics-otel, npm/ClawHub publishing prep) remains the most architecturally consequential plugin-packaging shift; 2026.4.27's Codex Computer Use + DeepInfra entrance + LINE inbound media path remain the watch → include-now promotion candidates; the seven 2026.5.7 quiet contract changes (Active Memory admin-scope #78863, Channels CLI restructuring #78456, Cron CLI status field #78701, native command owner enforcement #78864, auto-reply skill-tool authorization #78517, Codex pre-guardian PermissionRequest removal + allow-always session memory, Discord voice permission auditing) remain DOC11 readback-truth obligations. WATCH items from beta channel (pre-stable, not yet alignment-basis material): `/think default` + `/fast default` session-override clearing semantics intersect DOC11 native-command catalog + session-override floor truth; `openclaw.json` 0600 chmod restoration after suspicious-read backup recovery is a credential-handling hardening item touching EC Core secrets governance + DOC11 backup/recovery truth; `v2026.5.10-beta.6` contents not yet enumerated in detail. No new architectural surfaces today, no new schema obligations, no new Codex/route changes, no new plugin lifecycle changes. - Notes: WebSearch queries (`OpenClaw v2026 release notes site:github.com/openclaw/openclaw May 2026`, `"openclaw 2026.5.8" OR "openclaw 2026.5.10" OR "openclaw 2026.5.11" OR "openclaw 2026.5.12" stable release`, `openclaw latest release v2026.5 May 10 11 12 stable`) plus direct fetch of github.com/openclaw/openclaw/releases (top page) confirm latest stable tag is still `v2026.5.7` (2026-05-07T20:57:43Z, fixes-only, prerelease=false). New observation today: `v2026.5.10-beta.6` (2026-05-11, prerelease) on top of the already-known `v2026.5.9-beta.1` (2026-05-09, prerelease). No `v2026.5.8` of any form. Five-day stable pause is unusual against the prior 2026.4.27 → 2026.5.7 daily cadence and supports a strong recommendation to **actually start drafting** the DOC11 R15 R3 consolidated amendment during this pause rather than wait for further releases that would all need rolling together. Single-release roll-forwards have now lagged behind release cadence for fifteen consecutive days; the five-day pause does not narrow the gap, only halts its growth. R3 should fold 2026.4.27 (promote watch → include-now), 2026.4.29, 2026.5.2, 2026.5.3, 2026.5.3-1, 2026.5.4, 2026.5.5, 2026.5.6, AND 2026.5.7 in one consolidated pass — with the Codex-route saga folded as a single closed three-act item rather than three separate rows. Companion next-R of the running list (R2 → R3) carries the same logic. Dashboard updates today: `lastChecked` → 2026-05-12; `recommendation` rewritten to reflect today's no-change finding, fifth-consecutive-quiet-day status, fifteen-day cadence lag, new `v2026.5.10-beta.6` beta watch row alongside `v2026.5.9-beta.1`, and a strengthened call to begin drafting R3 during the quiet window; `releasesSinceLastCovered` / `latestReleaseChecked` / `latestReleaseCheckedDate` / `gapStatus` all preserved (unchanged); alignment-doc fields preserved per task constraints (alignment docs themselves are not modified). No 2026-05-11 nightly entry exists in this log — appears the tracker did not run yesterday; today's check stands as the first entry since 5/10 and so spans a two-day observation window rather than the usual one-day delta. --- ## 2026-05-10 10:42 — Nightly check - Alignment doc covers through: 2026.4.26 (2026-04-29, per DOC11 R15 R2 amendment proposal — newer than running list R2 at 2026-04-13 / 2026.4.12) - Latest stable release: 2026.5.7 (2026-05-07, published 20:57 UTC) — UNCHANGED from yesterday and the day before. Second consecutive quiet day. - Gap status: BEHIND (9 new releases — 2026.4.27, 2026.4.29, 2026.5.2, 2026.5.3, 2026.5.3-1, 2026.5.4, 2026.5.5, 2026.5.6, 2026.5.7; 2026.4.28 / 2026.4.30 / 2026.5.0 / 2026.5.1 / 2026.5.8 were not published as stable). UNCHANGED in composition from 2026-05-08 / 2026-05-09. - New since last covered: none today. No new stable OpenClaw release has been published since 2026.5.7 (2026-05-07 20:57 UTC). This is the **second consecutive quiet day** following yesterday's first quiet day in nearly two weeks. The daily-cadence streak that ran 2026.4.27 → 2026.5.7 has now paused for 48+ hours. WATCH-ONLY: a `v2026.5.9-beta.1` prerelease tag was published 2026-05-09 with chat-command additions (`/think default` and `/fast default` to clear session overrides and inherit configured/provider defaults) and a config/recovery hardening item (chmod restored `openclaw.json` back to owner-only `0600` on POSIX hosts after suspicious-read backup recovery, so a previously world-readable config mode cannot persist into a freshly restored credential-bearing config). Beta dist-tag only — does NOT promote into the stable alignment basis until cut as `v2026.5.9` (or higher) on the latest dist-tag. No `v2026.5.8` tag exists in any form (stable or beta) — the version-number sequence appears to be skipping 5.8 the way it skipped 4.28 / 4.30 / 5.0 / 5.1. - Most significant new items: none stable. The 9-release backlog narrative is unchanged from yesterday. Codex OAuth route saga remains closed in safe three-act form (2026.5.5 unsafe rewrite → 2026.5.6 revert → 2026.5.7 safe preserve/recover); DOC11 R15 R3 may treat the conditional Codex-route preservation rule as stable precedent for capability-bound default-runtime resolution. 2026.5.3's bundled file-transfer plugin + gateway invalid-config guards remain the most schema-consequential items; 2026.5.2's plugin externalization (ACPX, diagnostics-otel, npm/ClawHub publishing prep) remains the most architecturally consequential plugin-packaging shift; 2026.4.27's Codex Computer Use + DeepInfra entrance + LINE inbound media path remain the watch → include-now promotion candidates; the seven 2026.5.7 quiet contract changes (Active Memory admin-scope #78863, Channels CLI restructuring #78456, Cron CLI status field #78701, native command owner enforcement #78864, auto-reply skill-tool authorization #78517, Codex pre-guardian PermissionRequest removal + allow-always session memory, Discord voice permission auditing) remain DOC11 readback-truth obligations. WATCH item from beta channel: if `v2026.5.9-beta.1`'s `/think default` + `/fast default` chat-command additions land as stable, they intersect DOC11 native-command catalog + session-override clearing semantics (a quiet contract change for any spec that pins a session-override floor); the `openclaw.json` 0600 chmod restoration after suspicious-read backup recovery is a credential-handling hardening item touching EC Core secrets governance and DOC11 backup/recovery truth. Both items remain pre-stable and not yet alignment-basis material. No new architectural surfaces today, no new schema obligations, no new Codex/route changes, no new plugin lifecycle changes. - Notes: GitHub releases search (`OpenClaw v2026 release notes site:github.com/openclaw/openclaw`, `"openclaw 2026.5.8" OR "openclaw 2026.5.9" stable release`, and direct fetch of github.com/openclaw/openclaw/releases) confirms no `v2026.5.8` tag of any form, latest stable tag is still `v2026.5.7` (2026-05-07T20:57:43Z, fixes-only, prerelease=false), and `v2026.5.9-beta.1` was published 2026-05-09 as a prerelease (npm dist-tag beta — does not promote alignment basis). Single-release roll-forwards have now lagged behind release cadence for thirteen consecutive days; the two-day pause does not narrow the gap, only halts its growth. The two-day quiet window remains a useful drafting window for the DOC11 R15 R3 consolidated amendment that should fold 2026.4.27, 2026.4.29, 2026.5.2, 2026.5.3, 2026.5.3-1, 2026.5.4, 2026.5.5, 2026.5.6, AND 2026.5.7 — with the Codex-route saga folded as a single closed three-act item rather than three separate rows. Companion next-R of the running list (R2 → R3) carries the same logic. Dashboard updates today: `lastChecked` → 2026-05-10; `recommendation` rewritten to reflect today's no-change finding, the second-consecutive-quiet-day status, the thirteen-day cadence lag count, and the new beta-only 2026.5.9-beta.1 watch row; `releasesSinceLastCovered` / `latestReleaseChecked` / `latestReleaseCheckedDate` / `gapStatus` all preserved (unchanged); alignment-doc fields preserved per task constraints (alignment docs themselves are not modified). --- ## 2026-05-09 11:59 — Nightly check - Alignment doc covers through: 2026.4.26 (2026-04-29, per DOC11 R15 R2 amendment proposal — newer than running list R2 at 2026-04-13 / 2026.4.12) - Latest stable release: 2026.5.7 (2026-05-07, published 20:57 UTC) — UNCHANGED from yesterday - Gap status: BEHIND (9 new releases — 2026.4.27, 2026.4.29, 2026.5.2, 2026.5.3, 2026.5.3-1, 2026.5.4, 2026.5.5, 2026.5.6, 2026.5.7; 2026.4.28 / 2026.4.30 / 2026.5.0 / 2026.5.1 were not published as stable). UNCHANGED from 2026-05-08. - New since last covered: none today. No new stable OpenClaw release has been published since 2026.5.7 (2026-05-07 20:57 UTC). This is the first day in nearly two weeks without a new stable release — the daily-cadence streak that ran from 2026.4.27 through 2026.5.7 (with 2026.4.28 / 2026.4.30 / 2026.5.0 / 2026.5.1 skipped at the version-number level but every other day landing a stable cut) has paused. Gap composition is unchanged from yesterday. - Most significant new items: none. The 9-release backlog narrative is unchanged from yesterday: Codex OAuth route saga is closed in safe three-act form (2026.5.5 unsafe rewrite → 2026.5.6 revert → 2026.5.7 safe preserve/recover); 2026.5.3's bundled file-transfer plugin + gateway invalid-config guards remain the most schema-consequential items; 2026.5.2's plugin externalization (ACPX, diagnostics-otel, npm/ClawHub publishing prep) remains the most architecturally consequential plugin-packaging shift; 2026.4.27's Codex Computer Use + DeepInfra entrance + LINE inbound media path remain the watch → include-now promotion candidates; the seven 2026.5.7 quiet contract changes (Active Memory admin-scope #78863, Channels CLI restructuring #78456, Cron CLI status field #78701, native command owner enforcement #78864, auto-reply skill-tool authorization #78517, Codex pre-guardian PermissionRequest removal + allow-always session memory, Discord voice permission auditing) remain DOC11 readback-truth obligations. No new architectural surfaces today, no new schema obligations, no new Codex/route changes, no new plugin lifecycle changes. - Notes: GitHub releases search (`OpenClaw v2026.5.8 release notes site:github.com/openclaw/openclaw` and `"openclaw/openclaw" releases "2026.5.8" OR "2026.5.9" tag`) confirms no v2026.5.8 or v2026.5.9 tags exist; latest tag is still v2026.5.7 (2026-05-07T20:57:43Z, fixes-only, prerelease=false). npm registry and OpenClaw release-policy docs surface no newer version. Single-release roll-forwards have now lagged behind release cadence for twelve consecutive days; today's pause does not narrow the gap, only halts its growth. This is a useful drafting window for the DOC11 R15 R3 consolidated amendment that should fold 2026.4.27, 2026.4.29, 2026.5.2, 2026.5.3, 2026.5.3-1, 2026.5.4, 2026.5.5, 2026.5.6, AND 2026.5.7 — with the Codex-route saga folded as a single closed three-act item rather than three separate rows. Companion next-R of the running list (R2 → R3) carries the same logic. Dashboard updates today: lastChecked → 2026-05-09; recommendation rewritten to reflect today's no-change finding and the twelve-day lag count; releasesSinceLastCovered / latestReleaseChecked / latestReleaseCheckedDate / gapStatus all preserved (unchanged); alignment-doc fields preserved per task constraints (alignment docs themselves are not modified). --- ## 2026-05-08 10:47 — Nightly check - Alignment doc covers through: 2026.4.26 (2026-04-29, per DOC11 R15 R2 amendment proposal — newer than running list R2 at 2026-04-13 / 2026.4.12) - Latest stable release: 2026.5.7 (2026-05-07, published 20:57 UTC) - Gap status: BEHIND (9 new releases — 2026.4.27, 2026.4.29, 2026.5.2, 2026.5.3, 2026.5.3-1, 2026.5.4, 2026.5.5, 2026.5.6, 2026.5.7; 2026.4.28 / 2026.4.30 / 2026.5.0 / 2026.5.1 were not published as stable) - New since last covered: 2026.5.7 (2026-05-07 20:57 UTC) is new since the prior 2026-05-07 10:46 check — gap widened from BEHIND-by-8 to BEHIND-by-9. 2026.5.7 was published the day after 2026.5.6 (2026-05-06 17:51 UTC) but missed yesterday's pass because it landed ~10 hours after the nightly check window. - Most significant new items: 2026.5.7 is a fixes-only release with no new operative architectural surfaces but it lands the THIRD and final pass on the 2026.5.5/2026.5.6 Codex OAuth route saga: **doctor --fix now both PRESERVES working `openai-codex/*` PI routes AND RECOVERS 2026.5.5-rewritten `openai/*` GPT-5 routes back to `openai-codex/*` when only Codex OAuth auth is available** (fixes #78407, thanks @shakkernerd). This restores the safe form of the rule that 2026.5.6 reverted — yesterday's tracker said DOC11 R15 R3 should NOT canonize the abandoned 2026.5.5 logic; **today's update reverses that: the conditional Codex-route preservation rule is BACK in stable form**, but only as a forward-only PRESERVE/RECOVER rule (not the original aggressive REWRITE rule that broke OAuth-only GPT-5.5 setups). DOC11 R15 R3 may now treat the rule as stable precedent for capability-bound default-runtime resolution. Other architecturally noteworthy 2026.5.7 items — quiet contract changes DOC11 R15 should explicitly track: (1) **Active Memory** now requires admin scope for global memory toggles (#78863) — DOC11 OpenClaw memory feature controls + DOC15 Active Memory ownership now have an explicit admin/owner scope distinction; (2) **Channels CLI restructuring** — `openclaw channels list` is channel-only with `--all` flag for bundled/catalog channels rendering installed/configured/enabled state, while model auth/usage details moved to `openclaw models auth list`, `openclaw status`, and `openclaw models list` (#78456) — DOC11 channels truth + DOC10/DOC22 obligations; (3) **Cron CLI status field** added to `cron list --json` and `cron show --json` output (disabled/running/ok/error/skipped/idle) so external tooling does not have to reimplement cron status derivation (#78701); (4) **Native commands** now honor owner enforcement (#78864); (5) **Auto-reply** gates inline skill tool dispatch through before-tool-call authorization hooks (#78517) — DOC3 skills + DOC11 plugin hooks intersection; (6) **Codex approvals** stops installing pre-guardian native `PermissionRequest` hook by default so Codex's reviewer can approve safe commands before OpenClaw surfaces an approval; remembers `allow-always` decisions for identical Codex native `PermissionRequest` payloads within active session window; makes plugin approval requests validate/render their actual allowed decisions so Telegram and other native approval UIs cannot offer stale actions — affects Codex Computer Use governance / DOC11 Codex provider; (7) **Discord voice** permission auditing in `channels capabilities` and `channels status --probe` (including auto-join targets); extended default post-speech silence grace to 2.5s with `voice.captureSilenceGraceMs` for noisy Discord sessions — DOC10/DOC11 voice routing. Other notable 2026.5.7 fixes: Agents/context engine invalidates cached assembled context views when source history shrinks or assembly fails (#77968); Agents/subagents completed session-mode subagent registry rows honor `agents.defaults.subagents.archiveAfterMinutes` instead of hardcoded 5-minute TTL (#78263); Agents/compaction clamps compaction summary reserve tokens to each model's output limit (#54392); Plugins/install uses same absolute POSIX npm lifecycle shell for managed plugin install/rollback/repair/uninstall as staged package updates; Plugins/channel setup forwards `setChannelRuntime` from non-bundled external plugin setup entries (#77779); OpenAI supports `openai/chat-latest` as direct API-key model override; Tavily resolves dedicated `tavily_search`/`tavily_extract` tool credentials from active runtime config snapshot; Cron/doctor repairs persisted cron jobs whose `payload.model` was stored as `"default"`/`"null"`/blank/JSON null (#78549); Cron/isolated runs fail implicit announce delivery before model execution when `delivery.channel=last` has no previous route (#78608); Agent delivery reports `deliverySucceeded=false` when outbound delivery returns no adapter result (#78532); Gateway/sessions clears cached skills snapshots during `/new` and `sessions.reset` (#78873) and persists new generated transcript file when daily gateway-agent session rollover changes session id (#78607); Channels — Discord routes provider-prefixed `discord:channel:<id>` targets as channel sends; WhatsApp routes proactive phone-number sends through Baileys LID forward mappings (#67378); Telegram honors `accessGroup:*` sender allowlists (#78660), keeps polling watchdog tied to `getUpdates` liveness (#78422), parses provider ids containing dots in `/models` callback buttons (#38745); Model providers normalize APNG sniffed PNG uploads, preserve Gemini 3 tool-call thought-signature replay with fallback signatures, accept legacy `__env__:VAR` custom-provider keys, repair snake_case tool-call transcript sanitization. Carry-overs from prior days remain un-folded: 2026.5.6 Codex revert + plugin/debug-proxy/web-fetch header normalization; 2026.5.5 fixes-only stability work; 2026.5.4 Twilio voice bridge; 2026.5.3 file-transfer plugin; 2026.5.3-1 install-scanner hotfix; 2026.5.2 plugin externalization; 2026.4.29 active-run steering / memory wiki; 2026.4.27 Codex Computer Use / DeepInfra / LINE inbound media. Spec touchpoints across the 9-release backlog: DOC11, DOC4, DOC3, DOC15, DOC10, DOC20/22, DOC24, DOC25, EC Core. - Notes: GitHub releases API confirms v2026.5.7 is published as stable (prerelease=false; published_at 2026-05-07T20:57:43Z; release notes are fixes-only with no Features/New Surfaces section). No v2026.5.8 or higher tags exist yet. The 2026.5.5 → 2026.5.6 → 2026.5.7 Codex route trilogy is the architecturally most consequential development this week — it eliminates yesterday's narrative that DOC11 R15 R3 should NOT canonize the rule (the rule is now back in safe form). Single-release roll-forwards have now lagged behind release cadence for eleven consecutive days. Recommendation unchanged in structure but updated in substance: an R3 of the DOC11 R15 amendment proposal should fold 2026.4.27 (promote watch → include-now), 2026.4.29, 2026.5.2, 2026.5.3, 2026.5.3-1, 2026.5.4, 2026.5.5, 2026.5.6, AND 2026.5.7 in one consolidated pass — with the Codex-route saga folded as a single closed three-act narrative (5.5 unsafe rewrite → 5.6 revert → 5.7 safe preserve/recover) rather than three separate rows. Companion next-R of the running list. Dashboard latestReleaseChecked / latestReleaseCheckedDate / lastChecked / releasesSinceLastCovered (added 2026.5.7 entry; 2026.5.6 entry annotated with SUPERSEDED-by-2026.5.7 note) / recommendation all updated; alignment-doc fields preserved (alignment docs themselves are not modified per task constraints). --- ## 2026-05-07 10:46 — Nightly check - Alignment doc covers through: 2026.4.26 (2026-04-29, per DOC11 R15 R2 amendment proposal — newer than running list R2 at 2026-04-13 / 2026.4.12) - Latest stable release: 2026.5.6 (2026-05-06, published 17:51 UTC) - Gap status: BEHIND (8 new releases — 2026.4.27, 2026.4.29, 2026.5.2, 2026.5.3, 2026.5.3-1, 2026.5.4, 2026.5.5, 2026.5.6; 2026.4.28 / 2026.4.30 / 2026.5.0 / 2026.5.1 were not published as stable) - New since last covered: 2026.5.6 (2026-05-06 17:51 UTC) is new since the prior 2026-05-06 10:45 check — gap widened from BEHIND-by-7 to BEHIND-by-8. 2026.5.6 was published the same day as 2026.5.5 (~9 hours later) but missed yesterday's pass because it landed after the nightly check window. - Most significant new items: 2026.5.6 is a same-day fixes-only patch on top of 2026.5.5 with NO new operative architectural surfaces. The single most important change is a REVERT: 2026.5.6 backs out the 2026.5.5 `doctor --fix` Codex route repair that rewrote valid `openai-codex/*` ChatGPT/Codex OAuth routes to `openai/*` (fixes #78407). The 2026.5.5 logic could break OAuth-only GPT-5.5 setups or accidentally move users onto the OpenAI API-key route. Recovery command for users hit by 2026.5.5: `openclaw models set openai-codex/gpt-5.5 && openclaw config validate`. Recovery docs: https://docs.openclaw.ai/providers/openai#check-and-recover-codex-oauth-routing. **This directly invalidates the precedent that yesterday's recommendation pulled from 2026.5.5** — the conditional `agentRuntime.id: "codex"` iff Codex plugin installed/enabled/contributes codex harness/has usable OAuth, else `"pi"` rule that was flagged as a useful precedent for DOC11 capability-bound default-runtime resolution is now gone. DOC11 R15 R3 should NOT canonize it. Other 2026.5.6 fixes: Plugins/runtime fetch drops third-party symbol metadata from plain request header dictionaries before passing them into native `fetch` or `Headers` so SDK and guarded/proxy fetch paths do not reject otherwise valid plugin requests (fixes #77846, thanks @shakkernerd); Debug proxy normalizes captured fetch header dictionaries before replaying requests so symbol metadata from caller-owned header objects cannot make debug-proxy fetches fail; Web fetch bounds guarded dispatcher cleanup after request timeouts so timed-out fetches return tool errors instead of leaving Gateway tool lanes active (#78439, thanks @obviyus). None introduce schema-level obligations — pure plumbing/security hardening. Carry-overs from prior days remain un-folded: 2026.5.5 fixes-only stability work, 2026.5.4 Twilio realtime voice bridge, 2026.5.3 bundled file-transfer plugin + gateway invalid-config guards, 2026.5.2 plugin externalization, 2026.4.29 active-run steering / memory wiki, 2026.4.27 Codex Computer Use / DeepInfra / LINE inbound media. Likely spec touchpoints across the 8-release backlog: DOC11, DOC4, DOC3, DOC15, DOC10, DOC20/22, DOC24, DOC25, EC Core. - Notes: GitHub releases API and tags API (per_page=10) confirm v2026.5.6 is published as stable (prerelease=false; published_at 2026-05-06T17:51:03Z; release notes are fixes-only with no Features/New Surfaces section and contain a Codex OAuth recovery callout). No v2026.5.7 or higher tags exist yet. The 2026.5.5 → 2026.5.6 revert pair is the architecturally most consequential development since the 5/6 check because it eliminates a precedent that was actively guiding DOC11 R15 R3 thinking. Single-release roll-forwards have now lagged behind release cadence for ten consecutive days, with two same-day patches (2026.5.5 and 2026.5.6 both 5/6) underscoring the cost of single-release amendment cycles. Recommendation unchanged from yesterday in structure but updated in substance: an R3 of the DOC11 R15 amendment proposal should fold 2026.4.27 (promote watch → include-now), 2026.4.29, 2026.5.2, 2026.5.3, 2026.5.3-1, 2026.5.4, 2026.5.5, AND 2026.5.6 in one consolidated pass — with the explicit 2026.5.5 → 2026.5.6 revert pair noted so DOC11 does not lock in the abandoned Codex route-repair logic. Companion next-R of the running list. Dashboard latestReleaseChecked / latestReleaseCheckedDate / lastChecked / releasesSinceLastCovered / recommendation / 2026.5.5 summary (annotated with revert note) all updated; alignment-doc fields preserved (alignment docs themselves are not modified per task constraints). --- ## 2026-05-06 10:45 — Nightly check - Alignment doc covers through: 2026.4.26 (2026-04-29, per DOC11 R15 R2 amendment proposal — newer than running list R2 at 2026-04-13 / 2026.4.12) - Latest stable release: 2026.5.5 (2026-05-06, today, published 09:00 UTC) - Gap status: BEHIND (7 new releases — 2026.4.27, 2026.4.29, 2026.5.2, 2026.5.3, 2026.5.3-1, 2026.5.4, 2026.5.5; 2026.4.28 / 2026.4.30 / 2026.5.0 / 2026.5.1 were not published as stable) - New since last covered: 2026.5.5 (2026-05-06) is new since the prior 2026-05-05 check — gap widened from BEHIND-by-6 to BEHIND-by-7 - Most significant new items: 2026.5.5 (today) is a fixes-only release with no new operative architectural surfaces. Notable items: Doctor/sessions recovery-key migration for heartbeat-poisoned default main session store entries (doctor --fix can repair instances stuck on agent:main:main heartbeat history); Doctor/Codex legacy openai-codex/* → canonical openai/* route repair with conditional agentRuntime.id selection (codex when Codex plugin installed/enabled/has codex harness/has usable OAuth; else pi); Doctor/Gateway supervisor-restart-handoff visibility in openclaw doctor --deep and openclaw gateway status --deep; Plugins/update sync hardening for installed official npm + ClawHub plugins (including disabled or exact-pinned) with managed npm-root openclaw peer-link repair before installs and reassertion after shared-root npm installs/updates/uninstalls; Hooks/session-memory collision suffixes for fallback memory filenames + opt-in llmSlug for model-generated slugs + run reset-memory capture off command reply path; Agents/config removes ambiguous legacy "main" agent dir helper from runtime paths (default/session agent dirs now resolved through agents.list / agent-scope helpers; plugin SDK keeps deprecated compatibility export); CLI/sessions prunes unreferenced transcript / compaction-checkpoint / trajectory artifacts during normal sessions cleanup so gateway restart/crash orphans don't accumulate outside sessions.json; CLI/status + sessions CLI + Control UI/Sessions all show selected agent runtime/harness with Sessions-table runtime filter; Control UI compaction count compact "N Checkpoint(s)" disclosure with checkpoint history cards; Control UI restores session-memory and custom hook capture on explicit Control UI session creation; persisted assistant progress text stays visible after tool-use turn; TUI skips respawn wrapper for interactive launches and bounds session picker to recent rows; refuses to restore heartbeat sessions as remembered chat session; Gateway/OpenAI-compatible flushes initial chat-stream chunk and sends assistant-role SSE chunk as soon as headers accepted; Gateway/shutdown structured ShutdownResult warnings + cancel delayed post-ready maintenance during close + suppress maintenance/cron startup after quick restarts; Gateway/status avoids event-loop-degraded marking from CPU/utilization alone until sustained sampling window; xAI/Grok stops sending OpenAI-style reasoning effort to native Grok Responses models with bundled xAI thinking profile clamped to off (xai/grok-4.3 no longer fails Docker/Gateway with Invalid reasoning effort); Fireworks/Kimi exposed as thinking-off-only with K2.5/K2.6 pinned to thinking: disabled; Channels: Feishu hydrates missing native topic starter thread IDs before session routing; LINE rejects dmPolicy: "open" without wildcard allowFrom; Telegram/Codex deduplicates native Codex tool progress; Discord routes /steer through normal mention gate + shows live reasoning text in progress drafts + heartbeat ACK timeout measured from actual heartbeat send; Matrix retries approval delivery up to 3x with backoff; Slack preserves Socket Mode SDK error context; WhatsApp stops verified-stale local TUI clients to recover responsiveness; iOS pairing allows setup-code and manual ws:// for private LAN/.local gateways while keeping Tailscale/public on wss://; Docker/Gateway drops NET_RAW + NET_ADMIN capabilities and enables no-new-privileges in bundled docker-compose.yml; /status shows compact Gateway process uptime + host system uptime; Plugins/diagnostics actionable source-only TypeScript package warnings; Auth profiles avoid cooldown for format-level rejections; Exec approvals guarded copy fallback when Windows rejects rename-overwrite; Update dev-channel preflight lint made opt-in and constrained; Video generation provider-specific aspect-ratio/resolution hint normalization (720P → MiniMax 768P) + stop sending Google generateAudio on Gemini video. Carry-overs from prior days remain un-folded: 2026.5.4 Twilio realtime voice bridge for Meet/Voice Call, Codex audio-transcription manifest advertisement, Windows loopback 127.0.0.1 pin, plugins migration install hints; 2026.5.3 bundled file-transfer plugin (first-class native data-egress) + gateway/config invalid-state guards + lazy-loaded runtime/UI hot paths + onboarding install-hints + Codex-as-selectable-provider in Manual setup; 2026.5.2 plugin externalization (ACPX, diagnostics-otel, npm/ClawHub publishing prep), descriptor-first platform tool planner with cached descriptors, ClawPack metadata flow, threadBindings.spawnSessions consolidation, provider-native web search, Grok 4.3 xAI default; 2026.4.29 active-run steering / visible-reply enforcement / subagent routing metadata, people-aware memory wiki with provenance, per-conversation Active Memory filters, bounded REM preview diagnostics, tmux-backed gateway:watch, opt-in startup diagnostics timeline; 2026.4.27 Codex Computer Use, DeepInfra bundled provider, plugin lifecycle hardening, LINE inbound media path. Likely spec touchpoints across the 7-release backlog: DOC11, DOC4, DOC3, DOC15, DOC10, DOC20/22, DOC24, DOC25, EC Core. - Notes: GitHub releases API and tags API (per_page=20) confirm v2026.5.5 is published as stable (prerelease=false; published_at 2026-05-06T09:00:55Z; release notes are fixes-only with no Features/New Surfaces section). Two new beta tags (v2026.5.5-beta.1, v2026.5.5-beta.2) preceded the stable cut; no v2026.5.6 or higher tags exist yet. Single-release roll-forwards have not kept pace for nine consecutive days. None of the 2026.5.5 fixes introduce new schema-level obligations on the scale of 2026.5.3's file-transfer plugin or gateway invalid-config guards, but the Agents/config legacy "main" agent dir helper removal (resolution now via agents.list / agent-scope helpers; plugin SDK keeps deprecated compatibility export) is the kind of quiet contract change DOC11 R15 should explicitly track; the Doctor/Codex agentRuntime.id selection rule (codex iff Codex plugin installed + enabled + contributes codex harness + has usable OAuth, else pi) is a useful precedent for DOC11 capability-bound default-runtime resolution; CLI/sessions cleanup of unreferenced transcript/compaction-checkpoint/trajectory artifacts touches DOC11 sessions/checkpoint truth and EC durable-state retention assumptions. Recommendation unchanged from yesterday: an R3 of the DOC11 R15 amendment proposal should now fold 2026.4.27 (promote watch → include-now), 2026.4.29, 2026.5.2, 2026.5.3, 2026.5.3-1, 2026.5.4, AND 2026.5.5 in one consolidated pass, with a companion next-R of the running list. Dashboard latestReleaseChecked / latestReleaseCheckedDate / lastChecked / releasesSinceLastCovered / recommendation all updated; alignment-doc fields preserved (alignment docs themselves are not modified per task constraints). --- ## 2026-05-05 10:44 — Nightly check - Alignment doc covers through: 2026.4.26 (2026-04-29, per DOC11 R15 R2 amendment proposal — newer than running list R2 at 2026-04-13 / 2026.4.12) - Latest stable release: 2026.5.4 (2026-05-05, today) - Gap status: BEHIND (6 new releases — 2026.4.27, 2026.4.29, 2026.5.2, 2026.5.3, 2026.5.3-1, 2026.5.4; 2026.4.28 / 2026.4.30 / 2026.5.0 / 2026.5.1 were not published as stable) - New since last covered: 2026.5.3 (2026-05-04), 2026.5.3-1 hotfix (2026-05-04, beta-tag), and 2026.5.4 (2026-05-05) are new since the prior 2026-05-04 check — gap widened from BEHIND-by-3 to BEHIND-by-6 in ~24 hours - Most significant new items: 2026.5.4 (today) lands the Twilio realtime voice bridge for Google Meet / Voice Call (paced audio streaming, backpressure-aware buffering, barge-in queue clearing, no TwiML fallback during realtime speech); Codex audio transcription advertised in runtime/manifest metadata and routed to OpenAI transcription default rather than chat model IDs; plugins/migration emits catalog-backed install hints for missing referenced external plugins; gateway/Windows pins default loopback listener to 127.0.0.1 to avoid libuv dual-stack ::1 wedge; agents/performance threads resolved-workspace through BTW / compaction / embedded-run model generation; runtime + provider dependency refresh including Pi 0.73.0, ACPX adapters, OpenAI, Anthropic, Slack. 2026.5.3 (yesterday) lands a brand-new bundled file-transfer plugin (file_fetch / dir_list / dir_fetch / file_write agent tools, default-deny per-node path policy, symlink traversal refused by default, 16 MB byte ceiling per round-trip — first-class native data-egress surface), gateway/config invalid-state guards (no auto-restore at startup or hot reload; openclaw doctor --fix repairs), gateway/performance lazy-loading of plugin/runtime discovery / cron / schema / shutdown / sessions / model metadata until needed, channels/replies improvements (Discord status reactions + degraded transport reporting; WhatsApp Channel/Newsletter targets; Telegram, Feishu, Matrix, MS Teams, Slack delivery/recovery tightened), install/update recovery for broken macOS LaunchAgent upgrades, rejection of source-only plugin packages before runtime load, repair of stale Gateway/plugin state during updates and doctor runs, and onboarding install-hints with the external Codex plugin selectable in Manual setup. 2026.5.3-1 is a core npm hotfix on the beta dist-tag stopping the install scanner from blocking official bundled plugin packages when process.env access and normal API sends only appear in distant parts of the same compiled bundle. Carry-overs from prior days remain un-folded: 2026.5.2 plugin externalization (ACPX behind @openclaw/acpx, diagnostics OpenTelemetry behind @openclaw/diagnostics-otel, npm/ClawHub publishing prep), descriptor-first platform tool planner with cached plugin tool descriptors, ClawPack metadata flow, openclaw gateway restart --force / --wait <duration>, threadBindings.spawnSessions consolidation, provider-native web search, Grok 4.3 default xAI chat; 2026.4.29 active-run steering / visible-reply enforcement / spawned subagent routing metadata, people-aware memory wiki with provenance views, per-conversation Active Memory filters, bounded REM preview diagnostics, tmux-backed gateway:watch, opt-in startup diagnostics timeline; 2026.4.27 Codex Computer Use, DeepInfra bundled provider, plugin lifecycle hardening, LINE inbound media path. Likely spec touchpoints: DOC11, DOC4, DOC3, DOC15, DOC10, DOC20/22, DOC24, DOC25, EC Core. - Notes: GitHub releases API (per_page=15) confirms three new stable tags landed since the 2026-05-04 check: v2026.5.3 (published 2026-05-04, prerelease=false), v2026.5.3-1 (2026-05-04, prerelease=false but body notes core npm publish on the beta dist-tag), and v2026.5.4 (2026-05-05, prerelease=false). The bundled file-transfer plugin in 2026.5.3 is the single most significant new architectural surface in this batch — it introduces a first-class native data-egress capability (file_fetch / dir_list / dir_fetch / file_write across paired nodes) that DOC3 plugin catalog, DOC11 plugin lifecycle, DOC15/DOC25 ingestion safety, and EC durable-state policy will all need to classify and govern; the gateway/config invalid-state guards in 2026.5.3 also touch DOC11 runtime/config truth and EC durable-state recovery model. The Twilio realtime voice bridge in 2026.5.4 is the second-most-significant new item (DOC11 channels truth, DOC10 voice/audio routing, DOC22 voice surface, DOC24 KDA persisted-audio readback). Recommendation strengthened from yesterday: an R3 of the DOC11 R15 amendment proposal should now fold 2026.4.27 (promote watch → include-now), 2026.4.29, 2026.5.2, 2026.5.3, 2026.5.3-1, and 2026.5.4 in one consolidated pass, with a companion next-R of the running list — single-release roll-forwards are no longer keeping pace with the release cadence. Dashboard latestReleaseChecked / latestReleaseCheckedDate / lastChecked / releasesSinceLastCovered / recommendation all updated; alignment-doc fields preserved (alignment docs themselves are not modified per task constraints). --- ## 2026-05-04 10:43 — Nightly check - Alignment doc covers through: 2026.4.26 (2026-04-29, per DOC11 R15 R2 amendment proposal — newer than running list R2 at 2026-04-13 / 2026.4.12) - Latest stable release: 2026.5.2 (2026-05-02) - Gap status: BEHIND (3 new releases — 2026.4.27, 2026.4.29, 2026.5.2; 2026.4.28 / 2026.4.30 / 2026.5.0 / 2026.5.1 were not published as stable) - New since last covered: 2026.4.27, 2026.4.29, 2026.5.2 (unchanged from prior check) - Most significant new items: No change since 2026-05-03 entry — latest stable remains 2026.5.2 (~2 days old). Architecturally significant un-folded items still: 2026.5.2 plugin externalization (ACPX behind @openclaw/acpx, diagnostics OpenTelemetry behind @openclaw/diagnostics-otel, broad npm/ClawHub publishing prep), descriptor-first platform tool planner with cached plugin tool descriptors, ClawPack metadata flowing through install/onboarding/doctor-repair/channel-setup, openclaw gateway restart --force / --wait <duration>, threadBindings.spawnSessions consolidation with doctor --fix migration, provider-native web search reflecting configured provider, Grok 4.3 default xAI chat; 2026.4.29 active-run steering / visible-reply enforcement / spawned subagent routing metadata, people-aware memory wiki with provenance views, per-conversation Active Memory filters, bounded REM preview diagnostics, tmux-backed gateway:watch, opt-in startup diagnostics timeline; 2026.4.27 Codex Computer Use, DeepInfra bundled provider, plugin lifecycle hardening, LINE inbound media path. Likely spec touchpoints unchanged: DOC11, DOC4, DOC3, DOC15, DOC10, DOC20/22, DOC24, DOC25, EC Core. - Notes: WebSearch and GitHub release-tag scan confirm no new stable tag has been cut since 2026.5.2 (2026-05-02). Beta-only tags continuing on the 2026.5.2-beta.x line; nothing has crossed the stable cutover. Recommendation unchanged: DOC11 R15 R3 amendment proposal warranted to fold 2026.4.29 + 2026.5.2 and promote 2026.4.27 from watch to include-now, with companion next-R of the running list. Dashboard `lastChecked` advanced to 2026-05-04; all other OPENCLAW_RELEASE block fields preserved (latestReleaseChecked / latestReleaseCheckedDate / releasesSinceLastCovered / recommendation / gapStatus all still accurate from the 2026-05-03 update). --- ## 2026-05-03 10:42 — Nightly check - Alignment doc covers through: 2026.4.26 (2026-04-29, per DOC11 R15 R2 amendment proposal — newer than running list R2 at 2026-04-13 / 2026.4.12) - Latest stable release: 2026.5.2 (2026-05-02) - Gap status: BEHIND (3 new releases — 2026.4.27, 2026.4.29, 2026.5.2; 2026.4.28 / 2026.4.30 / 2026.5.0 / 2026.5.1 were not published as stable) - New since last covered: 2026.4.27, 2026.4.29, 2026.5.2 (2026.5.2 is new since the 2026-05-02 check — first 2026.5.x stable) - Most significant new items: 2026.5.2 lands plugin externalization architecture (ACPX behind @openclaw/acpx, diagnostics OpenTelemetry behind @openclaw/diagnostics-otel, broad npm/ClawHub publishing prep for Discord / MS Teams / WhatsApp / Voice Call / Memory LanceDB / Brave / Codex / Feishu / etc.); descriptor-first platform tool planner with cached plugin tool descriptors and runtime preload scoped to effective plugin IDs; ClawPack metadata flowing through install/onboarding/doctor-repair/channel-setup with digest verification; openclaw gateway restart --force / --wait <duration> plus active task run-ID logging before restart deferral; threadBindings.spawnSessions consolidation replacing split subagent/ACP toggles, with openclaw doctor --fix legacy-key migration; openclaw plugins list --json includes package dependency install state; provider-native web search now reflects configured provider rather than hard-coded Brave; Grok 4.3 added as default xAI chat model; OpenAI-compatible TTS extraBody passthrough; LM Studio reasoning metadata; broad messaging-channel fixes (WhatsApp Channel/Newsletter, Telegram topic commands, Discord, Slack threads, Signal groups, visible reply routing). 2026.4.29 carry-over (un-folded): active-run steering / visible-reply enforcement / spawned subagent routing metadata, people-aware memory wiki with provenance views and partial recall on timeout, per-conversation Active Memory filters, bounded REM preview diagnostics, heartbeat follow-up commitments, tmux-backed gateway:watch, opt-in startup diagnostics timeline. 2026.4.27 carry-over (un-folded): Codex Computer Use with fail-closed MCP checks, DeepInfra bundled provider, daemon PATH hardening, plugin install/inspect/uninstall lifecycle hardening, LINE inbound media persistence path. Likely spec touchpoints: DOC11, DOC4, DOC3, DOC15, DOC10, DOC20/22, DOC24, DOC25, EC Core. - Notes: 2026.5.2 is the first 2026.5.x stable. GitHub tags API (per_page=15) confirms version sequence skipped 2026.4.30 / 2026.5.0 / 2026.5.1 — those exist only as betas (2026.4.30-beta.1, 2026.5.1-beta.1/.2, 2026.5.2-beta.1/.2/.3 are pre-releases). 2026.5.2 introduces a meaningful architectural shift in plugin packaging (npm-first cutover, externalized ACPX/diagnostics-otel, ClawPack artifact metadata) that pre-empts several DOC11 R15 R2 plugin-runtime rows and broadens DOC4 plugin architecture obligations — strong candidate for promoting from a single-release roll-forward to a deeper amendment refresh. Recommendation: rather than continuing to defer R3, an R3 of the DOC11 R15 amendment proposal should fold in both 2026.4.29 and 2026.5.2 plus promote 2026.4.27 from watch to include-now, with a companion next-R of the running list. Dashboard latestReleaseChecked / latestReleaseCheckedDate / lastChecked / releasesSinceLastCovered / recommendation all updated; alignment-doc fields preserved (alignment docs themselves are not modified). --- ## 2026-05-02 — Nightly check - Alignment doc covers through: 2026.4.26 (2026-04-29, per DOC11 R15 R2 amendment proposal) - Latest stable release: 2026.4.29 (2026-04-29) - Gap status: BEHIND (2 new releases — 2026.4.27 and 2026.4.29; 2026.4.28 was not published) - New since last covered: 2026.4.27, 2026.4.29 (unchanged from prior check) - Most significant new items: No change since 2026-05-01 entry — latest stable remains 2026.4.29, and the most recent tag activity is the 2026.4.29-beta.4 raw-main pre-release. Carry-over architecturally significant items still un-folded: 2026.4.29's active-run steering / visible-reply enforcement / spawned subagent routing metadata, people-aware memory wiki with provenance views, per-conversation Active Memory filters, bounded REM preview diagnostics, heartbeat follow-up commitments, tmux-backed gateway:watch, opt-in startup diagnostics timeline; plus 2026.4.27's Codex Computer Use, DeepInfra bundled provider, plugin lifecycle hardening, LINE inbound media path. Likely spec touchpoints unchanged: DOC11, DOC15, DOC10, DOC20/22, DOC24, DOC25, EC Core. - Notes: Tag listing confirmed via GitHub tags API (per_page=20). No 2026.4.30 or 2026.5.x stable tag exists yet. Recommendation unchanged: next R of the running list and an R3 of the DOC11 R15 amendment proposal to fold 2026.4.29 in and promote 2026.4.27 from watch to include-now. Dashboard `lastChecked` advanced to 2026-05-02; no other OPENCLAW_RELEASE block fields needed to change. --- ## 2026-05-01 06:41 — Nightly check - Alignment doc covers through: 2026.4.26 (2026-04-29, per DOC11 R15 R2 amendment proposal — newer than running list R2 at 2026-04-13 / 2026.4.12) - Latest stable release: 2026.4.29 (2026-04-29) - Gap status: BEHIND (2 new releases — 2026.4.27 and 2026.4.29; 2026.4.28 was not published) - New since last covered: 2026.4.27, 2026.4.29 - Most significant new items: 2026.4.29 lands active-run steering / visible-reply enforcement / spawned subagent routing metadata, a people-aware memory wiki with provenance views and partial recall on timeout, per-conversation Active Memory filters, bounded REM preview diagnostics, heartbeat follow-up commitments, tmux-backed gateway:watch, opt-in startup diagnostics timeline, six new locales (fa/nl/vi/it/ar/th) + zh-TW docs glossaries, and a Yuanbao channel docs entrance. 2026.4.27 carry-over (still un-folded): Codex Computer Use with fail-closed MCP checks, DeepInfra bundled provider, daemon PATH hardening, plugin install/inspect/uninstall lifecycle hardening, LINE inbound media persistence path. Likely spec touchpoints: DOC11, DOC15, DOC10, DOC20/22, DOC24, DOC25, EC Core. - Notes: 2026.4.27 was already pre-classified in DOC11 R15 R2 as watch / stable_if_promoted, so its promotion to include-now is mechanical. 2026.4.29 is genuinely new and warrants a fresh R3 of the amendment proposal plus a new R of the running list. No 2026.4.28 release exists (404 on GitHub release tag), so version sequence skipped from .27 to .29. --- ## 2026-04-30 10:42 — Nightly check - Alignment doc covers through: 2026.4.26 (2026-04-29, per DOC11 R15 R2 amendment proposal — newer than running list R2 at 2026-04-13 / 2026.4.12) - Latest stable release: 2026.4.27 (2026-04-27) - Gap status: BEHIND (1 new release) - New since last covered: 2026.4.27 - Most significant new items: Codex Computer Use setup with fail-closed MCP checks; DeepInfra bundled provider with provider-owned onboarding (model discovery, media gen/edit, TTS, embeddings); daemon PATH hardening; plugin install/inspect/uninstall lifecycle improvements; LINE inbound media persisted under ~/.openclaw/media/inbound/. - Notes: DOC11 R15 R2 proposal already pre-classified 2026.4.27 as "watch / conditional include" with a stable_if_promoted disposition — its promotion to include-now is the main spec action. Single-release roll-forward, not a re-baseline. ---