- fix(code-mode): keep replay safety host-owned — Stack Beta-critical 1 of 3. - Base PR: main - Frozen construction base: c02e39345bf6946acd0635ac18732f4003dd45b5 - Head: 3bdc62f84d2c3f8f340934b2f2a8e4ba479615bc - Next:...
- perf: chat.history allocates ~12MB of garbage per request — Control UI's 2/sec poll saturates V8 old-space (recurring gateway OOM) — Summary chat.history performs, per request: a Buffer.alloc() of up to 12MB (session-utils.fs tail-read with maxBytes = max(maxHistoryBytes*2, 1MB)), a full .toString("utf-8") copy, and a...
- [Bug]: [ACP] Text before tool calls is lost in WeChat (intermediate messages dropped, not merged) — Bug type Behavior bug (incorrect output/state without crash) Beta release blocker No Summary When agent produces direct reply text between consecutive tool calls (write/edit/exec etc.) in the same...
- fix(exec): inherit tools.exec into new dashboard session defaults — Summary - New WebChat/Dashboard sessions do not seed execSecurity/execAsk, so reply-time overrides previously skipped persistent tools.exec and fell through toward deny. - Seed reply defaults from...
- readiness: adopt core runtime conditions — Related RFC: https://github.com/openclaw/rfcs/pull/33Depends on: https://github.com/openclaw/openclaw/pull/104018 at 2f131c6e220ProblemRFC 0018 provides an opt-in, bounded readiness facility, but...
- [Bug]: Inbound pipeline misclassifies mention-only short messages as assistant reasoning — Bug type Regression (worked before, now fails) Beta release blocker No Summary OpenClaw inbound pipeline misclassifies mention-only Feishu messages as assistant reasoning, replacing content with...
- improve(agents): benchmark concurrency hot paths — Closes #119568 What Problem This Solves OpenClaw had no focused, repeatable baseline for the core subagent concurrency lifecycle. Maintainers could not compare spawn fanout, admission accounting,...
- fix(bedrock): forward resolved maxTokens on reasoning-off requests — What Problem This Solves Amazon Bedrock Claude Opus 5 requests with reasoning: "off" omitted inferenceConfig.maxTokens even when the resolved model has a configured output cap, so output-heavy turns...
- feat(stepfun): default to step-3.7-flash — What Problem This Solves PR #88082 added step-3.7-flash to both direct StepFun catalogs (stepfun and stepfun-plan) but left the default model on step-3.5-flash. New onboarding runs and catalog...
- fix(gateway): flush throttled chat deltas after 150 ms — Fixes https://github.com/openclaw/openclaw/issues/119557. The Gateway currently retains a delta that arrives inside the 150 ms chat throttle until another qualifying event or terminal arrives....
- fix(qa): isolate Matrix wrong-key recovery accounts — What Problem This Solves Fixes an issue where the Matrix wrong-account recovery-key QA scenario used the shared observer account as its target and could reset that account's cross-signing or backup...
- [Bug]: On Chinese/legacy-codepage Windows, subprocess tool output is decoded with GBK instead of UTF-8 — On Windows systems whose system locale is Chinese (or any non-UTF-8 legacy codepage), tools that shell out to Python and capture stdout inherit the console/codepage encoding (cp936/GBK on zh-CN)....
- fix(sessions): decode Windows legacy-codepage exec output — Summary - reuse the shared strict-UTF-8-then-Windows-codepage streaming decoder for session-extension api.exec stdout and stderr - preserve the previous session executor contract for leading UTF-8...
- Legacy ackReaction migration is guarded per destination key, so a partly-set destination silently drops the scope half and still reports a clean move
- fix(telegram): show failed final slash-command responses — Related: #118545 Relates to gap finding: B3 (confirmed, medium) What Problem This Solves Telegram native slash commands could end silently when a non-final reply was intentionally suppressed and the...
- fix(auth-profiles): cool Anthropic OAuth profiles down until the real usage reset — What Problem This Solves Rate-limited Anthropic OAuth profiles were cooled down with the generic 30s to 5min backoff ladder, so OpenClaw retried within minutes and re-hit the same limit for the rest...
- [Bug]: Transient restart-required config write still restarts after revert to identical config — Summary A transient write to a restart-required config path (example: gateway.tools.allow), followed by a same-session revert that leaves openclaw.json byte-identical to the pre-change file, still...
- fix(agents): strip echoed base64 image payloads so photo-heavy turns no longer hit the per-turn output cap — > [AI-assisted] This PR was generated using Claude Code. > Maintainer repair and verification were performed with Codex. Closes #119445 Original pull request and proposal: @ml12580. Reported by...
- fix(ui): key ClawHub skill verdicts by canonical slug — What Problem This Solves The Control UI Skills page rendered Unavailable as the security badge for linked ClawHub skills even when skills.securityVerdicts returned ok=true, decision=pass,...
- [Bug]: cron pre-execution watchdog kills healthy runs — phase regression before_agent_reply→runtime_plugins re-arms a hardcoded 60s kill timer — Summary An embedded cron agent run that reaches before_agent_reply (execution stage) and then legitimately re-enters runtime_plugins — second turn, retry, subagent spawn — is reclassified...
- fix(memory-core): bound dreaming markdown reads — What Problem This Solves Fixes an issue where memory-core dreaming could buffer an oversized daily memory markdown file or DREAMS.md before updating its managed dreaming block, while avoiding a hard...
- improve(doctor): avoid repeated plugin metadata scans — Closes #119481 Problem An exact-main Kova diagnostic found 553 completed plugins.metadata.scan spans in each plugin-heavy gateway-performance record. The dominant Doctor process accounted for 493 of...
- [Bug]: Accepted WhatsApp turns can remain silent with no bounded liveness fallback — Bug type Behavior bug Is this a beta blocker? No Summary An accepted interactive WhatsApp turn can remain completely silent for tens of minutes even while the agent is active and emitting internal...
- fix(ollama): prevent credentials from appearing in embedding errors — or Related: # line below this comment. Required PR title: type: user-facing description Use a parenthesized scope only when it adds clarity: fix(auth): login redirect loops when session cookie is...
- fix(memory): allow retry after search timeout — Related: #93199 What Problem This Solves Fixes an issue where users retrying memory_search after its 15-second deadline would receive an embedding/provider error and remain blocked by a 60-second...
- fix(sessions): reject a transcript turn when the session id rotates mid-append (#119220) — What Problem This Solves persistSessionTranscriptTurn's unguarded path appends through the async SQLite transcript writer, which has no session-identity guard. When the session id rotates between...
- fix(ui): transcript search opens the selected historical message — Closes #118560 AI-assisted with Codex. I reviewed the implementation and validation results. What Problem This Solves Fixes an issue where users selecting a transcript-search result from before a...
- fix(gateway): cancel pending restart when config reverts to running baseline — Closes #119360 What Problem This Solves When a restart-required config change (A→B via gateway.tools.allow or similar) is deferred, the reload pipeline advances currentCompareConfig to B before the...
- feat(skills): preview proposals before approval — Closes #103871 Summary - add a read-only review action to skill_workshop - render create proposals as the complete canonical skill that would be applied and update proposals as unified diffs - bound...
- OpenClaw repo stats for 2026-08-05 — 385205 stars, 80971 forks, 3064 contributors
- OpenClaw sponsors (96 total) — 96 sponsors supporting OpenClaw
- fix(ci): authenticate skill updater pull requests (#3404) — Punchcard-Session: amber-orchard-valley-s4
- fix: resolve active skill across retained slug history (#3405)
- fix: enforce publisher skill slug invariant (#3383)
- fix(ui): use Carapace tokens for download trends (#3402)
- fix: allow Convex base64 upload digests (#3395)
- fix: emit plain Markdown in generated skill cards (#3385) — Fixes #3384. Generated skill cards now use plain Markdown instead of HTML-only line-break tags while retaining compatibility normalization for existing cards. Co-authored-by: Vyctor H. Brzezowski
- Add skill: ohad6k/ditto-profile
- mradermacher/Qwen3.5-9b-Opus-Openclaw-Distilled-i1-GGUF — text-generation
- mradermacher/Medina-Qwen3.5-27B-OpenClaw-Uncensored-i1-GGUF
- OpenClaw — Deploy your own personal AI assistant with OpenClaw on a DigitalOcean Droplet® server, a powerful open-source platform that runs entirely on your infrastructure. OpenClaw connects to the messaging...
- What agents learned in Synthetics' Last Cradle — On July 29, 2026, five OpenClaw agents sat down at Synthetics' Last Cradle and played for five hours...
- My OpenClaw Gateway Went Down at 3 AM. It Fixed Itself Before I Woke Up. — Something I've learned maintaining OpenClaw agents in production: the agent goes down when you least...
- Building Automated Workflows with ClawHub — If you've ever found yourself doing the same repetitive task over and over — scraping data every morning, copying files to specific folders, or manually gene...
- ClawHub Skill Spotlight: OpenClaw Automation Recipes — If you've ever wished your computer could just handle the boring stuff automatically, OpenClaw Automation Recipes is exactly what you need. This skill gives ...
- OpenClaw vs Hermes Agent: the Honest Comparison Nobody’s Given You Yet — Peter Steinberger built the first version of what became OpenClaw in about an hour.Continue reading on Towards AI »
- Donely Review: Managed AI Agents Still Need an Adult in the Loop — The main caveat with Donely is the same thing that makes it interesting: it is designed for persistent agents that can act, not merely…Continue reading on Medium »
- Best Mini PCs for OpenClaw in August 2026. Not an Outdated Article! — I Benchmarked 20 of Them So You Do Not Have To!Continue reading on Medium »
- Goose OSS — An open-source platform for seamless automation in engineering tasks, providing local execution and control. Customizable with any LLM for enhanced autonomy, enabling developers to efficiently handle...
- Maid (AI) — Maid is a cross-platform Flutter app for interfacing with GGUF / llama.cpp models locally, and with Ollama and OpenAI models remotely.
- Private GPT — Ask questions to your documents without an internet connection, using the power of LLMs. 100% private, no data leaves your execution environment at any point. You can ingest documents and ask...
- Apollo AI — Customizable interface for private, on-device AI model chatting, open source and closed source LLM connections, OpenRouter support, custom backend support, and unified access to language models...
- RT @k_adachi_01: KiroCrewが発表されました！ https://t.co/F9pqDEfaiH OpenClaw、Hermes AgentのようにSlackやDiscordから実行可能な常駐AIエージェントです。セッ — RT @k_adachi_01: KiroCrewが発表されました！ https://t.co/F9pqDEfaiH OpenClaw、Hermes AgentのようにSlackやDiscordから実行可能な常駐AIエージェントです。セッションをまたいでメモリが永続化され、K…
- RT @platzi: Acabamos de actualizar uno de los cursos más populares de Platzi de los últimos meses. 🌟Curso de OpenClaw 🌟
- RT @SirisCrypt: Did you know Openclaw and Hermes biggest competitor has twice the active monthly users as they do combin
- 🤖 What if your own 𝗔𝗜 𝗔𝘀𝘀𝗶𝘀𝘁𝗮𝗻𝘁 could build products while you sleep? 🚀 That's exactly what #OpenClaw did for #TranslaS
- RT @Hey_GraceWilson: 🚀 16 Must-Know AI GitHub Repos (2026) Save this list for your next AI project 👇 1. OpenClaw – Per
- RT @techradar: Rethinking defense in the wake of OpenClaw attacks https://t.co/KLbkPJejZa
- RT @Hey_GraceWilson: 🚀 16 Must-Know AI GitHub Repos (2026) Save this list for your next AI project 👇 1. OpenClaw – Per
- @rezoundous This is why i still unfortunately use openclaw with adaptive thinking.
- @dcuthbert Can't wait for someone's OpenClaw instance to start vomiting emails to the list. 😑 Many fond memories of Bug
- Rethinking defense in the wake of OpenClaw attacks https://t.co/KLbkPJejZa
- @Bollish check @Parad0x_Labs ca:8EeDdvCRmFAzVD4takkBrNNwkeUTUQh4MscRK5Fzpump NULLA → Local AI .null → Ownable identity — @Bollish check @Parad0x_Labs ca:8EeDdvCRmFAzVD4takkBrNNwkeUTUQh4MscRK5Fzpump NULLA → Local AI .null → Ownable identity DNA x402 → Agent payments Dark Null → Privacy Liquefy → Data + verifiable audit...
- RT @NicolasChaillan: Everyone kept telling me AI was "impressive." Impressive is a magic trick. Impressive is a demo.
- お気に入りのLFM2.5に兄貴分「LFM2.5-2.6B」が登場キタコレ！🔥 【半年前にYouTubeで騒いだあのLFM2.5、今度は「働くAI」になって帰ってきたのでメモシェア🙌】 覚えてる方いますかね。今年1月、ダイチ @LxGtUG — お気に入りのLFM2.5に兄貴分「LFM2.5-2.6B」が登場キタコレ！🔥 【半年前にYouTubeで騒いだあのLFM2.5、今度は「働くAI」になって帰ってきたのでメモシェア🙌】 覚えてる方いますかね。今年1月、ダイチ @LxGtUGtlRSh8yXW さんと撮影したYouTube動画で検証した、 Liquid https://t.co/vvwOconpa3...
- RT @platacrypto: To bring this highly technical piece of information in some simpler terms: @rei_labs is working to sol
- RT @CrypSaf: It’s been 8 days since I bought my Mac mini for OpenClaw 🦞 Here’s what I’m bleeding on: 🩸 > Mac mini -
- @BuildingWaves @openclaw Yeah I’d be down
- @LittleBill902 @theo Yeah it's what happens when you let openclaw run your X account lol
- Aujourd’hui, je me lance dans l’entrepreneuriat. Je suis développeur depuis près de dix ans. Depuis quelques mois, je t
- 【Microsoft Scout（スカウト）】とは 2026年6月発表。 デスクトップ版GUIのAIエージェントアプリ。 Scout＝ ModelはGPTやOpus等で選択できる ＋ ハーネスは OpenClaw ＋ 契約はGi — 【Microsoft Scout（スカウト）】とは 2026年6月発表。 デスクトップ版GUIのAIエージェントアプリ。 Scout＝ ModelはGPTやOpus等で選択できる ＋ ハーネスは OpenClaw ＋ 契約はGitHub Copilotの従量課金 ＋ 統制は Microsoft EntraID ＋ Purview ＋ Intune https://t.co/5PrfVNRxWF
- Every PDF parser fails on something. Tables break. Multi-column layouts get scrambled. Formulas vanish. Scanned pages co