- fix(test): type hidden launcher env assertion
- fix(bedrock): stop replaying stale signed thinking from completed turns — What Problem This Solves On Amazon Bedrock (Converse), the message converter replays every historical signed thinking block unconditionally. Over a long, tool-heavy session these signed reasoning...
- perf(agents): index displaced tool results
- fix(memory-lancedb): surface memory ID in memory_recall content — Summary Fixes the recall→forget handoff gap where the model can list a memory via memory_recall but cannot delete it via memory_forget(memoryId=...) because the recall output only shows numbered...
- fix(ollama): bound model-discovery JSON response reads — What Problem This Solves extensions/ollama/src/provider-models.ts discovers Ollama models by calling /api/tags (fetchOllamaModels) and /api/show (queryOllamaModelShowInfo), and parsed both responses...
- fix(agents): normalize mangled tool names and IDs from OpenAI-compati… — Summary - Problem: Some OpenAI-compatible providers (e.g. Kimi) send tool call IDs like functions.exec:0 as functions exec:0 (space instead of dot) and tool names like exec as functions exec. Tool...
- fix(copilot): preserve compaction metadata — What Problem This Solves The Copilot harness already bridges compaction into the generic runtime, but successful SDK compactions were dropping the canonical compaction metadata that sibling harnesses...
- fix(ui): rename chat sessions from the picker — What Problem This Solves - Chat/sidebar session picker had no inline rename capability. Users had to navigate to the Sessions table to rename a session, breaking their chat workflow. - Fix: Add...
- feat(hooks): add agent and tool lifecycle boundaries — Why Agent execution lacked complete lifecycle boundaries around thinking, response generation, and tool execution, reducing traceability for observability and policy systems. This PR adds those...
- fix(sandbox): materialize protected skill mount roots — What Problem This Solves Fixes #94425. A first-launch rw sandbox can currently skip the read-only overlays for /workspace/skills and /workspace/.agents/skills when the corresponding host directories...
- fix: assistant reply lost between compaction summary and first kept user in successor transcript — Summary buildSuccessorEntries drops all summarized assistant replies from the successor transcript after compaction rotation, leaving a one-sided conversational gap. Preserve the last assistant...
- fix(workboard): filter archived cards from CLI list output — Summary The openclaw workboard list CLI command shows archived (soft-deleted) cards by default, inconsistent with the workboard_list agent tool and /workboard list slash command which both hide them....
- fix(net): pass NO_PROXY explicitly to undici EnvHttpProxyAgent (fixes #95998) — What Problem This Solves Issue #95998 reports that when HTTPS_PROXY is set globally, the qqbot plugin's COS chunked upload is routed through the proxy even when NO_PROXY includes .myqcloud.com. The...
- Improve model call telemetry — What Problem This Solves OpenClaw model-call traces made the current-turn input easy to see, but the real prompt burden could be split across model input messages, system instructions, and tool...
- feat(huggingface): add text-to-image generation via hf-inference Inference Providers route — Summary Re-opens #63487 in a much smaller, focused form per @steipete's review feedback. Adds text-to-image only to the existing huggingface plugin, routed through the official Hugging Face Inference...
- ci: add release QA profile evidence — Summary: - add a release-check QA job that runs the taxonomy-backed release profile with pnpm openclaw qa run --qa-profile release - wire the job into the existing qa/all release-check paths,...
- fix(memory-core): index memory path in FTS text for filename queries (fixes #94102) — What Problem This Solves - memory_search FTS indexed only chunk body text. Queries for date-like filename stems (e.g. 2026-06-17-1649) returned unrelated files with textScore: 0 even when the target...
- fix(memory-core): index memory path in FTS text for filename queries (fixes #94102) — What Problem This Solves - memory_search FTS indexed only chunk body text. Queries for date-like filename stems (e.g. 2026-06-17-1649) returned unrelated files with textScore: 0 even when the target...
- test: prefer shared temp dir helpers in auto-reply and install-fallback tests — What Problem This Solves ~100+ test files still use raw fs.mkdtemp/fs.mkdtempSync instead of the shared temp directory helpers in src/test-helpers/temp-dir.ts. This causes: 1. Disk leaks — temp dirs...
- test: prefer shared temp dir helpers in auto-reply and install-fallback tests — What Problem This Solves ~100+ test files still use raw fs.mkdtemp/fs.mkdtempSync instead of the shared temp directory helpers in src/test-helpers/temp-dir.ts. This causes: 1. Disk leaks — temp dirs...
- Autofix: add PR review autofix pipeline + Windows daemon — Summary Adds a PR review autofix pipeline that uses the Claude Agent SDK subscription to address review comments automatically, plus a Windows background daemon that runs the loop + supervises the...
- fix #89231: [Bug]: Windows installer-created scheduled task launches gateway.cmd with visible console — should use windowless launcher — Summary Fixes #89231. Windows gateway service installs now request and actually apply the existing hidden launcher by default. The generated gateway service environment sets...
- [Bug]: Windows installer-created scheduled task launches gateway.cmd with visible console — should use windowless launcher — Summary On Windows, the OpenClaw scheduled task OpenClaw Gateway (created at install time and bound to LogonTrigger) runs gateway.cmd directly. Because the action is cmd.exe → node.exe, a visible...
- [Bug]: Setup fails at `pair-operator` step: `interop=false` in wsl.conf blocks Windows → WSL localhost connectivity — Bug type Behavior bug (incorrect output/state without crash) Beta release blocker No Summary The OpenClaw Companion installer consistently fails during the pair-operator step across all three...
- fix(agents): sessions_send delivery.status now accurately reflects actual delivery state — Fixes #96020 The sessions_send tool always returned delivery.status as "pending" even after the message was successfully sent to the target session. This caused downstream pendingFinalDelivery...
- Prevent repeated byte-triggered compactions — Summary - prevent repeated byte-triggered preflight compactions for the same active transcript when a previous byte-triggered compaction did not materially reduce the transcript below the configured...
- Deny same-provider cross-context message sends by default — Summary Deny same-provider cross-context message tool sends by default. A tool call bound to one conversation can still reply to the current conversation, but sending/uploading into another...
- fix(openshell): prevent recursively nested directories in mirror uploads — Summary - Root cause: OpenShell sandbox upload preserves the source directory's basename as a subdirectory inside the remote target when uploading a named path (OpenShell PRs #952 and #1028,...
- fix(agents): detect peer/conversation IDs from --bind values via channel plugin hook — What Problem This Solves agents add --bind "feishu:oc_xxx" stores the value after the colon in match.accountId. For Feishu, IDs starting with oc_ are chat/group identifiers (chat_id), not channel...
- Gateway 'plugins.slots.memory: plugin not found' for an extensions/ plugin present in plugins list — 2026.6.9 regression — Summary On 2026.6.9, the Gateway rejects a valid config at startup with for a memory-slot plugin that lives in ~/.openclaw/extensions// (cloned + built, valid openclaw.plugin.json,...
- OpenClaw repo stats for 2026-06-23 — 380047 stars, 79583 forks, 2656 contributors
- OpenClaw sponsors (109 total) — 109 sponsors supporting OpenClaw
- feat(catalog): publish verified organization skills feed
- fix(catalog): version and harden hosted plugin feed (#2806)
- feat(catalog): publish hosted OpenClaw plugin feed (#2805) — feat(schema): define hosted catalog feed contract feat(catalog): publish cached hosted plugin feed feat(catalog): add feed publication and edge delivery fix(catalog): recheck live official publisher...
- feat: publish security dataset from live export
- Publish security dataset from live Convex export (#2800) — feat(cli): restore skill sync command feat: publish security dataset from live export
- fix(web): keep staff skill pages hydrated (#2798)
- Revert "docs: document security dataset export workflow" — This reverts commit 8841ac777196020e5a43d2ccf902874adfb76dec.
- docs: document security dataset export workflow
- fix(web): recall publishers outside browse window in search (#2790) — fix: recall publishers outside top install window in search Publisher search only scanned the top 500 by installs and dropped empty profiles, so handles like vincentkoc never appeared even when the...
- test: stabilize production menu smoke — Stabilizes production menu smoke by clicking exact header nav links and avoiding mobile drawer transition races between SPA navigations.\n\nVerification:\n- PLAYWRIGHT_BASE_URL=https://clawhub.ai...
- test: update prod smoke skill canonical route — Updates production HTTP smoke to expect the owner-qualified skill canonical URL.\n\nVerification:\n- bun run test:e2e:prod-http\n- git diff --check
- feat: add owner-qualified catalog routes — Adds canonical owner-qualified publisher, skill, and plugin routes while preserving legacy redirects.\n\nIncludes API, CLI, docs, and user-facing copy updates for //skills/ and //plugins/.\n\nMerged...
- fix(ci): report testbox hydration failures (#2787)
- fix(schema): expand reserved catalog topic slugs (#2791) — Block additional trust, curation, brand, and channel topic labels that publishers could use to imply platform endorsement.
- feat: reclaim deleted org handles — Adds an admin-only deleted-org handle reclaim path and hard-delete cleanup for empty deleted org tombstones. Also makes local-auth publish flows resilient to cold local Convex startup timeouts.
- fix: align home app card icons (#2788)
- feat(web): polish skill and plugin detail pages (#2780) — fix: polish skill detail hero metadata fix: improve skill readme presentation chore: outline skill detail structure fix: narrow detail page container fix: compact skill sidebar on detail pages fix:...
- feat(cli): restore skill sync command — Restores the ClawHub CLI sync command as a one-way skill publish workflow without install/download telemetry updates.
- fix: route banned users to account page (#2785)
- ykarout/Qwen3.5-9b-Opus-Openclaw-Distilled-GGUF — text-generation
- mradermacher/OpenCLAW-SEED-135M-GGUF — text-generation
- OpenClaw on DigitalOcean — 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...
- Anthropic says OpenClaw-style Claude CLI usage is allowed again
- Running a local AI agent on an 8GB Jetson Orin Nano (2026) — A practical look at running an always-on AI agent on an 8GB Jetson Orin Nano — what works, what doesn’t, and how to skip the setup pain.
- My OpenClaw Cron Broke and Fixed Itself Before I Noticed — Last Thursday I woke up to a Telegram alert: "Cron self-repair: 1 repair made." I hadn't triggered...
- My OpenClaw Security Audit Hung Every Night — A 3-Line Bash Fix and What It Taught Me About Sandboxing Agents
- Cómo construí un agente de IA que trabaja 24/7 (y lo que aprendí en el camino)
- Standardizing Enterprise Agentic Workflows: Implementing MCP Connectors in OpenClaw Frameworks — 1. Introduction: The Agent-to-Data Connectivity ChallengeContinue reading on Medium »
- Your AI Should Know You by Now — Prevail.sh is an AI that thinks in life domains, captures what you actually care about, and convenes a panel of models for the decisions…Continue reading on Medium »
- The Rise of Autonomous AI Agents: OpenClaw vs. PicoClaw — A deep dive into the self-hosted “Jarvis” ecosystems transforming developer workflows, smart homes, and edge computingContinue reading on Future Nexus »
- The Rise of Operational AI: What Open claw and MCP Signal for Digital Transformation Leaders — The next phase of enterprise AI will not be defined by systems that merely answer questions, but by systems that can participate in work.Continue reading on Medium »
- arXiv — 에이전트 생성형 LLM을 위한 집단 스킬 트리 탐색 기술, OpenClaw-Skill — 1. 한눈에 보는 핵심 요약 (Executive Summary)Continue reading on Medium »
- [OpenClaw #2] Setting Up OpenClaw and Running My First Agent — From Installation to First ConversationContinue reading on Medium »
- Manus — AI agent bridges thoughts and actions, excelling in work and life tasks like personalized travel, stock analysis, insurance comparisons, and supplier sourcing, autonomously completing tasks and...
- Open Claw- MS/Anthropic
- How do I view server logs from PaperclipAI being run via "npx paperclipai run"? Seeking assistance debugging "internal server error" from curl
- OpenClaw threads filling up with agents that died overnight when the Claude/Gemini OAuth terms shifted. The memory, the
- Prosus launches ToqanClaw, a no-code AI platform for over 5 million restaurant owners and entrepreneurs, and Zapia, an A
- RT @AYi_AInotes: Claude Code、OpenClaw、Hermes、Codex，一个比一个能干， 但 2026 年了，这批最先进的 Agent 还集体卡在同一件小事上： 上不了网，查推特要付费 API，读小红书卡登录
- RT @ZypherHQ: Why would anyone use OpenClaw over Codex? That's the question I genuinely can't answer, and I'd love for s
- @ZypherHQ Codex gives you answers. OpenClaw gives you control
- @steipete Can you elaborate about "have other agendas"? Thanks for bringing OpenClaw, it's awesome :)
- QQ邮箱正式开启专为AI智能体（Agent）设计的专属邮箱「Agently Mail」内测。 新邮箱与用户个人或企业邮箱的数据完全隔离，允许AI使用独立身份进行邮件收发、搜索和附件处理，避免AI因误删、错发或过度读取对用户个人隐私造成安全
- RT @TheAhmadOsman: If you’re using Ollama switch to llama.cpp If you’re still on OpenClaw switch to Hermes These are b
- RT @birdabo: 🚨TECH DRAMA: OpenClaw vs Hermes Agent. so… the OpenClaw founder just posted about how strong their week wa
- @IndieDevFire @openclaw i need to ✉️ u but cant
- @ZypherHQ OpenClaw or Codex? All in on OSS moon.
- RT @OpenBuildxyz: Join now! OpenClaw Summer Builder Bootcamp provides 8 weeks of infrastructure support, mentorship, a — RT @OpenBuildxyz: Join now! OpenClaw Summer Builder Bootcamp provides 8 weeks of infrastructure support, mentorship, and pathway to real…
- RT @wiliam23820a: Best 15 accounts to follow in AI: @karpathy = LLMs king @steipete = built openclaw @gregisenberg
- Setting up your openclaw or Hermes agent on Telegram? First, go build a bot. Get the API key from BotFather. Write the
- @Atticus_of_Oz @kimmonismus Alright, bit of a long answer here, but here you go. I can't speak for everybody, and I don'
- @LitianaPaeniu You can’t say that the ppl behind @OpenClaw aren’t passionate.
- @knutmt @SnillAI Interesting, so its like openclaw? What do you think about openclaw but like a personal assistant for b
- RT @steipete: People here discussing what happened with OpenClaw. The hype died down. We improved quality and grew a te
- RT @DataChaz: never ever leaving this app 🍿🍿🍿😎 The beef between Hermes and OpenClaw escalated quickly https://t.co/XOuL
- 考虑到openclaw在重型任务的交付表现上远不如codex，决定调整工作模式为： OpenClaw（网关 + 记忆 + Orchestrator） + Codex（任务执行 / Heavy Lifter）