OpenClaw Newsletter - 2026-06-27
OpenClaw Newsletter
Saturday, June 27, 2026Top Stories
- test: add sandbox escape prevention tests for skill loader — Summary Problem: The loadSkillsFromDirSafe and readSkillFrontmatterSafe functions in the skill loader lacked comprehensive automated test coverage for sandbox escape vectors. While the implementation...
- fix(media-understanding): append actionable install hint when media provider is missing — What Problem This Solves When a user has configured a media provider plugin that is not installed (e.g. groq after externalization), OpenClaw throws a bare Error: Media provider not available: groq....
- DeepSeek cache hit rate <10% after 6.x upgrade - boundary-aware caching breaks prefix matching — Environment - OpenClaw version: 2026.6.8 (upgraded from 2026.5.28) - Model: DeepSeek V4 Flash / V4 Pro - API type: openai-completions - Provider: Direct DeepSeek API (api.deepseek.com) Problem After...
- fix(bedrock): apply model maxTokens to adaptive-thinking and Fable 5 requests — What Problem This Solves Fixes #97176 — Amazon Bedrock Claude models using adaptive thinking (Opus 4.8, Fable 5) are hard-capped at 4096 output tokens. At higher thinking levels the thinking block...
- [Bug]: Mid-turn context overflow precheck aborts after successful truncation without re-verifying or submitting prompt — Bug type Regression (worked before, now fails) Beta release blocker No Summary The mid-turn context overflow precheck routes to truncate_tool_results_only, truncation reports success (handled=true,...
- fix(ollama): send tool_call arguments as strings on OpenAI-compatible path — Closes #96441 What Problem This Solves Fixes an issue where users on any Ollama Cloud model (:cloud, e.g. ollama/glm-5.2:cloud, glm-5.1:cloud, MiniMax M3 Cloud) would have multi-turn tool use break...
- fix(plugin-sdk): remove stale reasoning envelope when DeepSeek V4 wrapper emits native effort (#97196) — Related to #97196. What Problem This Solves OpenRouter-routed DeepSeek V4 requests fail with HTTP 400 because the chat-completions builder emits reasoning: { effort } (OpenRouter format) and the...
- fix(diagnostic): avoid V8 dictionary mode degradation in stream hot paths (#91588) — Fix #91588 — Gateway OOM crash: RSS grows from 350MB to 15.5GB over 2-3 days of normal use. Root Cause Two hot paths in the diagnostic event code cause V8 hidden class degradation (dictionary mode)...
- fix(codex): wait for turn completion after post-tool handoff — What Problem This Solves Codex Harness could terminate a turn too early after a tool handoff. The incorrect boundary was not the content of assistant text itself, but the lifecycle signal used by the...
- feat: add tokenomics plugin for local-first LLM spend accounting — What Problem This Solves Operators running OpenClaw have no built-in way to see what their model usage actually costs. Token usage is emitted as diagnostics, but there is no durable record of...
- fix: avoid DeepSeek-native thinking on OpenRouter V4 — Fixes #97196 What Problem This Solves Fixes an issue where users running OpenRouter-routed DeepSeek V4 models could hit a 400 response because OpenClaw sent both OpenRouter reasoning controls and...
- fix: gate sandbox write tools on writable workspace access — Fixes #65316. This narrows sandboxed coding-tool availability so write, edit, and apply_patch are exposed only when the sandbox has an effective writable workspace target. The prior path/media issues...
- fix: zhipu silentOverflow + cron watchdog stage misclassification — What Problem This Solves 1. Zhipu provider GLM models falsely flagged as overflow-prone Dennis (zhipu/GLM-5.2) transcripts were wedging because isSilentOverflowProneModel in agent-settings.ts uses...
- Add Control UI session kind filter — Summary - Adds a Control UI chat-header session kind filter for main/direct, group, subagent, dreaming, cron, and other sessions. - Keeps filtering UI-local so sessions.list params and gateway...
- Add Discord response watchdog — Fresh minimal branch superseding stale PR #73232 and auto-closed PR #73238. Task: 22d72cfc — Add Discord response watchdog for silent agent runs. Iris verification before PR creation: focused...
- fix(openai-responses): bound SSE response reads via buildGuardedModelFetch — What Problem This Solves The openai-responses provider creates an OpenAI SDK client without a custom fetch option, so all SSE streaming responses from the OpenAI API are read without any byte cap. A...
- feat(session-message-events) socket.drain — Summary 🦞 A resilient, deterministic path ensuring downstream systems immediately recognize session evictions, safely bypassing pub/sub pipeline delays and connection speed limits. It also firmly...
- fix: resolve response prefix template variables in abort replies — Summary Template variables are not populated when user issues early stop commands. - Problem: When a user sends a "stop" command (e.g. on WhatsApp), the abort reply displays unresolved template...
- Unhandled Playwright assertion error in CRSession._onMessage crashes Gateway — Bug Description An unhandled Assertion error from Playwright's Chrome DevTools Protocol (CDP) session management crashes the entire Gateway process. The error is not caught by OpenClaw, causing a...
- security: gateway public network hardening — Summary Implements comprehensive security hardening for OpenClaw Gateway when exposed to public networks (e.g. https://openclaw.allegro.earth/). The existing security mechanisms (non-loopback auth...
- [Bug]: Concurrent allow-always approvals silently lose allowlist entries (last-write-wins race in exec-approvals.json)
- [Bug]: Discord routing / mention-gating issue in OpenClaw — Bug type Regression (worked before, now fails) Summary Summary I found a Discord preflight / mention-gating bug where message handling is too permissive in some agent-routing cases. The issue is not...
- feat(backup): add exclude patterns, --smart-exclude, and protected path guards [ai] — Summary - Problem: openclaw backup create has no way to exclude files/directories, causing (1) bloated archives from venvs/, models/, logs/, completions/ and (2) silent backup stalls on installations...
- fix: serialize outbound deliveries per channel+recipient — Summary When multiple sessions (main + sub-agent) send messages to the same recipient concurrently, replies can arrive out of order because each session's delivery call races independently on the...
- Backup: add encrypted snapshot backup flow — Summary - Problem: OpenClaw had local backup export/verify, but no low-friction folder-backed snapshot flow for encrypted backups plus workspace mirroring. - Why it matters: users need a simple...
- fix(gateway): use account-scoped reload for channel account changes — Summary Use account-scoped hot reload for config changes under: - channels..accounts..* instead of always restarting the entire channel. What changed - added account-scoped restart planning to...
- Gateway: account-scoped channel config changes restart the entire channel and can disrupt active traffic — Summary When a config change only affects channels..accounts..*, the gateway currently plans a full channel restart instead of an account-scoped restart. This is visible in the reload pipeline even...
- [Feature]: Automatic Output Sanitization for Sensitive Data — Summary Add automatic output sanitization to prevent accidental leakage of sensitive information (API keys, tokens, passwords, PII) in agent responses, regardless of whether the requester is a tester...
- [Bug]: Sandbox prune does not clean up workspace directory — Bug type Behavior bug (incorrect output/state without crash) Summary Sandbox prune removes Docker containers but leaves workspace directories behind, causing disk space leak over time Steps to...
- OpenClaw repo stats for 2026-06-27 — 380668 stars, 79754 forks, 2676 contributors
- OpenClaw sponsors (110 total) — 110 sponsors supporting OpenClaw
- fix: update codex worker cli pin
- fix: reduce static scanner false positives — Reduce false positives in ClawHub static moderation scans while preserving dangerous child_process detection and secret literal coverage.
- fix: publish security snapshots to live dataset — Publish the moving security snapshot workflow to OpenClaw/clawhub-security-signals-live using a single latest split, with a maintained live dataset card.
- fix: raise catalog skills feed cap (#2912)
- fix: cap skills feed publication
- docs: mark feed schema as OpenClaw contract
- fix: restore catalog feed schema v1 (#2909)
- ci: default plugin inspector scans to dry run (#2908)
- fix: allow high-shard security dataset exports
- fix: export latest security dataset skills only
- fix(ci): give skill card worker Convex fallback — Fix Skill Card Worker to use the same production Convex fallback URL as the security worker and dataset snapshot workflows.\n\nEvidence:\n- Failed run 28269571442 had CONVEX_URL empty in all worker...
- fix: render static audit findings (#2904)
- fix: record security dataset hf commit
- fix: publish security dataset from existing shards
- style(test): format package publish workflow test
- fix(ci): hard cap security dataset fanout (#2901)
- test(cli): preserve literal lockfile version type
- test(ci): update artifact extraction assertion
- fix(ci): reject unsafe catalog feed expiry
- fix(ci): pin skill card worker dependencies
- openclaw-cli 2026.6.10 on Homebrew — Your own personal AI assistant
- openclaw/openclaw on Docker Hub — Official openclaw/openclaw docker images.
- 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...
- OpenClaw – Moltbot Renamed Again
- OpenClaw is what Apple intelligence should have been
- 4 Safety Boundaries Your AI Agent Needs Before Production (And How to Wire Them) — Most AI agents go into production with zero guardrails. Here's the 4-boundary framework I run on every agent before it touches real data — and the exact code to implement each one.
- 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.
- OpenClaw’s memory: a deep dive — So, I have been using openclaw, along with many of my peers, and we were impressed by how good it was at storing memories. We would often…Continue reading on Medium »
- OpenClaw vs Hermes: two bets on what an AI agent should be — I’ve run enough agent frameworks to know that the marketing rarely survives contact with a terminal. So when two open-source projects…Continue reading on Medium »
- Parallel Free Search Lands in OpenClaw: A Zero-Key Web Search MCP, Tested Against Tavily — OpenClaw just shipped parallel-free — a no-account, no-API-key web search provider built on Parallel's MCP.Continue reading on Level Up Coding »
- How AgentLine solves calling for AI Agents. — AgentLine is solving one of the most important problems that AI agents face today. Agents can do a lot of tasks now because the models…Continue reading on Medium »
- Build a Private Photo Assistant on Telegram with OpenClaw + MiniCPM-V 4.6 — Most AI assistants can read text, write code, and automate workflows. But the moment a user sends a photo — a receipt, a whiteboard…Continue 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...
- MiroThinker — MiroThinker is an open-source search agent model, built for tool-augmented reasoning and real-world information seeking, aiming to match the deep research experience of OpenAI Deep Research and...
- 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 @AYi_AInotes: 现在用AI做视频可以跟喝水一样简单，不需要再付个700多块的剪映SVIP， 装这6个2026 年最顶的插件和skills就够了， 链接直接丢给你的AI Agent（Claude Code、Cursor、
- openclaw on $MUMU —signals all pointing up $MUMU ┃ mc $1.5M ┃ vol $11.2K ┃ liq $95.4K ┃ tgt $4.5M ┃ +191% Was Made U
- my openclaw scan on $熊猫 just confirmed what i was thinking $熊猫 ├ mc $121.1K ├ vol $11.4K ├ liq $28.7K └ tgt $369.2
- my openclaw scan on $ALGOPUB just confirmed what i was thinking $ALGOPUB ▁▂▃▅▇ → $377.2K mc $88.5K · vol $11.8K · liq
- openclaw called $$OSMOSIS perfectly last time. running the scan again $$OSMOSIS · health 🟩🟩🟩🟩🟩🟩🟩🟨⬜⬜ mc $13.6K · vol $
- openclaw called $WNTV bullish. i'm listening $WNTV · health 🟥⬜⬜⬜⬜⬜⬜⬜⬜⬜ mc $4.2M · vol $11.2K · liq $166.8K target $1
- openclaw called it on $ATTENTION again. the model doesn't miss > ATTENTION MKT CAP $232.2K 24H VOL $15.6K
- my openclaw analyst flagged $nomnom as a buy. noted 🦞 $nomnom ┃ mc $326.3K ┃ vol $13.7K ┃ liq $113.1K ┃ tgt $969.0K ┃
- been tracking $BREW on openclaw for a week. forecast holding strong $BREW current → $1.4M target → $4.2M vol
- HAHA its fun with organic sales handled by openclaw! try out https://t.co/htO8xkasKA!! for all your homeopathy needs! ht
- just ran $PEPONK through openclaw. risk is low, upside is real $PEPONK ├ mc $14.2K ├ vol $12.3K ├ liq $9.3K └ tgt
- $tbh signals are clean on openclaw. this one's on my radar 🦞 > TBH MKT CAP $635.2K 24H VOL $10.9K LIQU
- openclaw been right on $HypurrClaw twice already. trusting the scan $HypurrClaw current → $81.1K target → $543.
- $SON momentum score just came in hot on openclaw 🔥 $SON · health 🟩🟩🟩🟩🟩⬜⬜⬜⬜⬜ mc $81.6K · vol $15.7K · liq $25.8K targ
- been watching $CODEC through openclaw, momentum is building $CODEC ▁▂▃▅▇ → $8.0M mc $2.9M · vol $21.8K · liq $274.8K
- openclaw been right on $MUSHU twice already. trusting the scan $MUSHU ▁▂▃▅▇ → $1.0M mc $404.4K · vol $11.7K · liq $16
- my openclaw analyst flagged $MOMO as a buy. noted 🦞 > MOMO MKT CAP $138.0K 24H VOL $15.3K LIQUIDITY $3
- openclaw confirmed what the chart was already telling me on $AIRA $AIRA | MC $71.2K → $525.2K | Vol $14.5K | +638% W
- been tracking $POKEX on openclaw for a week. forecast holding strong $POKEX ▁▂▃▅▇ → $137.3K mc $27.4K · vol $11.4K ·
- RT @runware: Deploy OpenClaw free in five minutes OpenClaw running in five minutes. One click deploys a secure instance
Trending on X
- RT @AYi_AInotes: 现在用AI做视频可以跟喝水一样简单，不需要再付个700多块的剪映SVIP， 装这6个2026 年最顶的插件和skills就够了， 链接直接丢给你的AI Agent（Claude Code、Cursor、
- openclaw on $MUMU —signals all pointing up $MUMU ┃ mc $1.5M ┃ vol $11.2K ┃ liq $95.4K ┃ tgt $4.5M ┃ +191% Was Made U
- my openclaw scan on $熊猫 just confirmed what i was thinking $熊猫 ├ mc $121.1K ├ vol $11.4K ├ liq $28.7K └ tgt $369.2
- my openclaw scan on $ALGOPUB just confirmed what i was thinking $ALGOPUB ▁▂▃▅▇ → $377.2K mc $88.5K · vol $11.8K · liq
- openclaw called $$OSMOSIS perfectly last time. running the scan again $$OSMOSIS · health 🟩🟩🟩🟩🟩🟩🟩🟨⬜⬜ mc $13.6K · vol $
- openclaw called $WNTV bullish. i'm listening $WNTV · health 🟥⬜⬜⬜⬜⬜⬜⬜⬜⬜ mc $4.2M · vol $11.2K · liq $166.8K target $1
- openclaw called it on $ATTENTION again. the model doesn't miss > ATTENTION MKT CAP $232.2K 24H VOL $15.6K
- my openclaw analyst flagged $nomnom as a buy. noted 🦞 $nomnom ┃ mc $326.3K ┃ vol $13.7K ┃ liq $113.1K ┃ tgt $969.0K ┃
- been tracking $BREW on openclaw for a week. forecast holding strong $BREW current → $1.4M target → $4.2M vol
- HAHA its fun with organic sales handled by openclaw! try out https://t.co/htO8xkasKA!! for all your homeopathy needs! ht
- just ran $PEPONK through openclaw. risk is low, upside is real $PEPONK ├ mc $14.2K ├ vol $12.3K ├ liq $9.3K └ tgt
- $tbh signals are clean on openclaw. this one's on my radar 🦞 > TBH MKT CAP $635.2K 24H VOL $10.9K LIQU
- openclaw been right on $HypurrClaw twice already. trusting the scan $HypurrClaw current → $81.1K target → $543.
- $SON momentum score just came in hot on openclaw 🔥 $SON · health 🟩🟩🟩🟩🟩⬜⬜⬜⬜⬜ mc $81.6K · vol $15.7K · liq $25.8K targ
- been watching $CODEC through openclaw, momentum is building $CODEC ▁▂▃▅▇ → $8.0M mc $2.9M · vol $21.8K · liq $274.8K
- openclaw been right on $MUSHU twice already. trusting the scan $MUSHU ▁▂▃▅▇ → $1.0M mc $404.4K · vol $11.7K · liq $16
- my openclaw analyst flagged $MOMO as a buy. noted 🦞 > MOMO MKT CAP $138.0K 24H VOL $15.3K LIQUIDITY $3
- openclaw confirmed what the chart was already telling me on $AIRA $AIRA | MC $71.2K → $525.2K | Vol $14.5K | +638% W
- been tracking $POKEX on openclaw for a week. forecast holding strong $POKEX ▁▂▃▅▇ → $137.3K mc $27.4K · vol $11.4K ·
- RT @runware: Deploy OpenClaw free in five minutes OpenClaw running in five minutes. One click deploys a secure instance
Releases
- OpenClaw repo stats for 2026-06-27 — 380668 stars, 79754 forks, 2676 contributors
- openclaw-cli 2026.6.10 on Homebrew — Your own personal AI assistant
- openclaw/openclaw on Docker Hub — Official openclaw/openclaw docker images.
Community
- test: add sandbox escape prevention tests for skill loader — Summary Problem: The loadSkillsFromDirSafe and readSkillFrontmatterSafe functions in the skill loader lacked comprehensive automated test coverage for sandbox escape vectors. While the implementation...
- fix(media-understanding): append actionable install hint when media provider is missing — What Problem This Solves When a user has configured a media provider plugin that is not installed (e.g. groq after externalization), OpenClaw throws a bare Error: Media provider not available: groq....
- DeepSeek cache hit rate <10% after 6.x upgrade - boundary-aware caching breaks prefix matching — Environment - OpenClaw version: 2026.6.8 (upgraded from 2026.5.28) - Model: DeepSeek V4 Flash / V4 Pro - API type: openai-completions - Provider: Direct DeepSeek API (api.deepseek.com) Problem After...
- fix(bedrock): apply model maxTokens to adaptive-thinking and Fable 5 requests — What Problem This Solves Fixes #97176 — Amazon Bedrock Claude models using adaptive thinking (Opus 4.8, Fable 5) are hard-capped at 4096 output tokens. At higher thinking levels the thinking block...
- [Bug]: Mid-turn context overflow precheck aborts after successful truncation without re-verifying or submitting prompt — Bug type Regression (worked before, now fails) Beta release blocker No Summary The mid-turn context overflow precheck routes to truncate_tool_results_only, truncation reports success (handled=true,...
- fix(ollama): send tool_call arguments as strings on OpenAI-compatible path — Closes #96441 What Problem This Solves Fixes an issue where users on any Ollama Cloud model (:cloud, e.g. ollama/glm-5.2:cloud, glm-5.1:cloud, MiniMax M3 Cloud) would have multi-turn tool use break...
- fix(plugin-sdk): remove stale reasoning envelope when DeepSeek V4 wrapper emits native effort (#97196) — Related to #97196. What Problem This Solves OpenRouter-routed DeepSeek V4 requests fail with HTTP 400 because the chat-completions builder emits reasoning: { effort } (OpenRouter format) and the...
- fix(diagnostic): avoid V8 dictionary mode degradation in stream hot paths (#91588) — Fix #91588 — Gateway OOM crash: RSS grows from 350MB to 15.5GB over 2-3 days of normal use. Root Cause Two hot paths in the diagnostic event code cause V8 hidden class degradation (dictionary mode)...
- fix(codex): wait for turn completion after post-tool handoff — What Problem This Solves Codex Harness could terminate a turn too early after a tool handoff. The incorrect boundary was not the content of assistant text itself, but the lifecycle signal used by the...
- feat: add tokenomics plugin for local-first LLM spend accounting — What Problem This Solves Operators running OpenClaw have no built-in way to see what their model usage actually costs. Token usage is emitted as diagnostics, but there is no durable record of...
- fix: avoid DeepSeek-native thinking on OpenRouter V4 — Fixes #97196 What Problem This Solves Fixes an issue where users running OpenRouter-routed DeepSeek V4 models could hit a 400 response because OpenClaw sent both OpenRouter reasoning controls and...
- fix: gate sandbox write tools on writable workspace access — Fixes #65316. This narrows sandboxed coding-tool availability so write, edit, and apply_patch are exposed only when the sandbox has an effective writable workspace target. The prior path/media issues...
- fix: zhipu silentOverflow + cron watchdog stage misclassification — What Problem This Solves 1. Zhipu provider GLM models falsely flagged as overflow-prone Dennis (zhipu/GLM-5.2) transcripts were wedging because isSilentOverflowProneModel in agent-settings.ts uses...
- Add Control UI session kind filter — Summary - Adds a Control UI chat-header session kind filter for main/direct, group, subagent, dreaming, cron, and other sessions. - Keeps filtering UI-local so sessions.list params and gateway...
- Add Discord response watchdog — Fresh minimal branch superseding stale PR #73232 and auto-closed PR #73238. Task: 22d72cfc — Add Discord response watchdog for silent agent runs. Iris verification before PR creation: focused...
- fix(openai-responses): bound SSE response reads via buildGuardedModelFetch — What Problem This Solves The openai-responses provider creates an OpenAI SDK client without a custom fetch option, so all SSE streaming responses from the OpenAI API are read without any byte cap. A...
- feat(session-message-events) socket.drain — Summary 🦞 A resilient, deterministic path ensuring downstream systems immediately recognize session evictions, safely bypassing pub/sub pipeline delays and connection speed limits. It also firmly...
- fix: resolve response prefix template variables in abort replies — Summary Template variables are not populated when user issues early stop commands. - Problem: When a user sends a "stop" command (e.g. on WhatsApp), the abort reply displays unresolved template...
- Unhandled Playwright assertion error in CRSession._onMessage crashes Gateway — Bug Description An unhandled Assertion error from Playwright's Chrome DevTools Protocol (CDP) session management crashes the entire Gateway process. The error is not caught by OpenClaw, causing a...
- security: gateway public network hardening — Summary Implements comprehensive security hardening for OpenClaw Gateway when exposed to public networks (e.g. https://openclaw.allegro.earth/). The existing security mechanisms (non-loopback auth...
- [Bug]: Concurrent allow-always approvals silently lose allowlist entries (last-write-wins race in exec-approvals.json)
- [Bug]: Discord routing / mention-gating issue in OpenClaw — Bug type Regression (worked before, now fails) Summary Summary I found a Discord preflight / mention-gating bug where message handling is too permissive in some agent-routing cases. The issue is not...
- feat(backup): add exclude patterns, --smart-exclude, and protected path guards [ai] — Summary - Problem: openclaw backup create has no way to exclude files/directories, causing (1) bloated archives from venvs/, models/, logs/, completions/ and (2) silent backup stalls on installations...
- fix: serialize outbound deliveries per channel+recipient — Summary When multiple sessions (main + sub-agent) send messages to the same recipient concurrently, replies can arrive out of order because each session's delivery call races independently on the...
- Backup: add encrypted snapshot backup flow — Summary - Problem: OpenClaw had local backup export/verify, but no low-friction folder-backed snapshot flow for encrypted backups plus workspace mirroring. - Why it matters: users need a simple...
- fix(gateway): use account-scoped reload for channel account changes — Summary Use account-scoped hot reload for config changes under: - channels..accounts..* instead of always restarting the entire channel. What changed - added account-scoped restart planning to...
- Gateway: account-scoped channel config changes restart the entire channel and can disrupt active traffic — Summary When a config change only affects channels..accounts..*, the gateway currently plans a full channel restart instead of an account-scoped restart. This is visible in the reload pipeline even...
- [Feature]: Automatic Output Sanitization for Sensitive Data — Summary Add automatic output sanitization to prevent accidental leakage of sensitive information (API keys, tokens, passwords, PII) in agent responses, regardless of whether the requester is a tester...
- [Bug]: Sandbox prune does not clean up workspace directory — Bug type Behavior bug (incorrect output/state without crash) Summary Sandbox prune removes Docker containers but leaves workspace directories behind, causing disk space leak over time Steps to...
- OpenClaw sponsors (110 total) — 110 sponsors supporting OpenClaw
- fix: update codex worker cli pin
- fix: reduce static scanner false positives — Reduce false positives in ClawHub static moderation scans while preserving dangerous child_process detection and secret literal coverage.
- fix: publish security snapshots to live dataset — Publish the moving security snapshot workflow to OpenClaw/clawhub-security-signals-live using a single latest split, with a maintained live dataset card.
- fix: raise catalog skills feed cap (#2912)
- fix: cap skills feed publication
- docs: mark feed schema as OpenClaw contract
- fix: restore catalog feed schema v1 (#2909)
- ci: default plugin inspector scans to dry run (#2908)
- fix: allow high-shard security dataset exports
- fix: export latest security dataset skills only
- fix(ci): give skill card worker Convex fallback — Fix Skill Card Worker to use the same production Convex fallback URL as the security worker and dataset snapshot workflows.\n\nEvidence:\n- Failed run 28269571442 had CONVEX_URL empty in all worker...
- fix: render static audit findings (#2904)
- fix: record security dataset hf commit
- fix: publish security dataset from existing shards
- style(test): format package publish workflow test
- fix(ci): hard cap security dataset fanout (#2901)
- test(cli): preserve literal lockfile version type
- test(ci): update artifact extraction assertion
- fix(ci): reject unsafe catalog feed expiry
- fix(ci): pin skill card worker dependencies
News
- 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...
- OpenClaw – Moltbot Renamed Again
- OpenClaw is what Apple intelligence should have been
- 4 Safety Boundaries Your AI Agent Needs Before Production (And How to Wire Them) — Most AI agents go into production with zero guardrails. Here's the 4-boundary framework I run on every agent before it touches real data — and the exact code to implement each one.
- 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.
- OpenClaw’s memory: a deep dive — So, I have been using openclaw, along with many of my peers, and we were impressed by how good it was at storing memories. We would often…Continue reading on Medium »
- OpenClaw vs Hermes: two bets on what an AI agent should be — I’ve run enough agent frameworks to know that the marketing rarely survives contact with a terminal. So when two open-source projects…Continue reading on Medium »
- Parallel Free Search Lands in OpenClaw: A Zero-Key Web Search MCP, Tested Against Tavily — OpenClaw just shipped parallel-free — a no-account, no-API-key web search provider built on Parallel's MCP.Continue reading on Level Up Coding »
- How AgentLine solves calling for AI Agents. — AgentLine is solving one of the most important problems that AI agents face today. Agents can do a lot of tasks now because the models…Continue reading on Medium »
- Build a Private Photo Assistant on Telegram with OpenClaw + MiniCPM-V 4.6 — Most AI assistants can read text, write code, and automate workflows. But the moment a user sends a photo — a receipt, a whiteboard…Continue 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...
- MiroThinker — MiroThinker is an open-source search agent model, built for tool-augmented reasoning and real-world information seeking, aiming to match the deep research experience of OpenAI Deep Research and...
- 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...
Don't miss what's next. Subscribe to Openclaw Newsletter: